Overall benchmark leader
Claude Fable 5.1
87.99 / 100 · Anthropic
The model selection desk
AI model leaderboard.
Compare model runs across seven suites from an independent benchmark. Start with a task, inspect the tradeoffs, and follow every number back to its source.
This snapshot was checked more than seven days ago. Review the source release for changes before making a selection.
coding suite leader
Claude Fable 5
98.30 / 100 · Anthropic
research suite leader
Claude Opus 5
88.90 / 100 · Anthropic
49 model runs
Filter keeps the full-board rank · select up to 3 models
Overall is the equal-weight mean of all seven suite scores. Models with missing suites remain visible but receive no overall rank.
No matching models. Try a shorter name or reset the provider filter.
| Rank | Model / measured configuration | Overall score | USD / benchmark task | Coverage | Compare |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.1Anthropic · effort: max | 87.99 | $4.484 | 7/7suite scores | |
| 2 | Claude Opus 5.5Anthropic · effort: max | 86.53 | $3.626 | 7/7suite scores | |
| 3 | GPT-6 AstraOpenAI · effort: max | 86.14 | $3.277 | 7/7suite scores | |
| 4 | GPT-6.1 SolOpenAI · effort: max | 85.41 | $0.620 | 7/7suite scores | |
| 5 | Gemini 4 ArgonGoogle · effort: high | 84.10 | $1.516 | 7/7suite scores | |
| 6 | Claude Fable 5Anthropic · effort: max | 83.61 | $5.313 | 7/7suite scores | |
| 7 | Claude Sonnet 5.5Anthropic · effort: max | 82.46 | $2.899 | 7/7suite scores | |
| 8 | Claude Opus 5Anthropic · effort: max | 82.09 | $3.654 | 7/7suite scores | |
| 9 | GPT-6 SolOpenAI · effort: max | 81.91 | $0.865 | 7/7suite scores | |
| 10 | DeepSeek V4.1 Flash 1017DeepSeek · effort: max · source-reported open weights | 79.20 | $0.131 | 7/7suite scores | |
| 11 | Muse Wonder 1.0Meta · effort: max | 78.10 | $1.409 | 7/7suite scores | |
| 12 | DeepSeek V4.1 FlashDeepSeek · effort: max · source-reported open weights | 78.04 | $0.167 | 7/7suite scores | |
| 13 | Claude Opus 4.8Anthropic · effort: max | 76.77 | $3.812 | 7/7suite scores | |
| 14 | GPT-5.6 SolOpenAI · effort: max | 76.26 | $1.835 | 7/7suite scores | |
| 15 | Claude Haiku 5.5Anthropic · effort: max | 75.64 | $0.619 | 7/7suite scores | |
| 16 | GPT-5.6 TerraOpenAI · effort: max | 74.99 | $1.206 | 7/7suite scores | |
| 17 | Grok 4.6xAI · effort: xhigh | 74.94 | $0.928 | 7/7suite scores | |
| 18 | MiMo V2.6 ProXiaomi · effort: high · source-reported open weights | 74.87 | $0.125 | 7/7suite scores | |
| 19 | Muse Spark 1.3Meta · effort: max | 74.76 | $0.794 | 7/7suite scores | |
| 20 | Grok 4.7xAI · effort: xhigh | 74.09 | $1.991 | 7/7suite scores | |
| 21 | MiniMax M3.1 FlashMiniMax · effort: max · source-reported open weights | 73.09 | $0.106 | 7/7suite scores | |
| 22 | Claude Sonnet 5Anthropic · effort: max | 72.99 | $2.595 | 7/7suite scores | |
| 23 | GLM-5.3 FlashZ.AI · effort: max · source-reported open weights | 70.61 | $0.108 | 7/7suite scores | |
| 24 | Grok 4.5xAI · effort: xhigh | 70.06 | $0.519 | 7/7suite scores | |
| 25 | GPT-6 LunaOpenAI · effort: max | 69.40 | $0.088 | 7/7suite scores | |
| 26 | Qwen3.8 FlashAlibaba · effort: max · source-reported open weights | 68.79 | $0.135 | 7/7suite scores | |
| 27 | GPT-5.5OpenAI · effort: xhigh | 68.70 | $2.204 | 7/7suite scores | |
| 28 | Gemini 3.8 FlashGoogle · effort: high | 67.34 | $0.789 | 7/7suite scores | |
| 29 | Gemini 3.7 FlashGoogle · effort: high | 67.19 | $0.617 | 7/7suite scores | |
| 30 | GPT-5.6 LunaOpenAI · effort: max | 66.09 | $0.223 | 7/7suite scores | |
| 31 | Claude Sonnet 4.6Anthropic · effort: max | 65.09 | $1.684 | 7/7suite scores | |
| 32 | Qwen3.8 MaxAlibaba · effort: max · source-reported open weights | 64.24 | $1.024 | 7/7suite scores | |
| 33 | DeepSeek V4 Flash VisionDeepSeek · effort: max · source-reported open weights | 63.63 | $0.141 | 7/7suite scores | |
| 34 | Kimi K3Moonshot AI · effort: max · source-reported open weights | 63.39 | $1.196 | 7/7suite scores | |
| 35 | Qwen3.8 27BAlibaba · effort: max · source-reported open weights | 62.66 | $0.519 | 7/7suite scores | |
| 36 | Muse Spark 1.2Meta · effort: xhigh | 62.40 | $0.580 | 7/7suite scores | |
| 37 | Step 5 (Preview)StepFun · effort: max · source-reported open weights | 62.01 | $0.750 | 7/7suite scores | |
| 38 | GLM-5.3Z.AI · effort: max · source-reported open weights | 60.74 | — | 7/7suite scores | |
| 39 | Muse Spark OpenMeta · effort: max · source-reported open weights | 57.57 | $0.488 | 7/7suite scores | |
| 40 | Beam 1.0Reflection AI · effort: max · source-reported open weights | 55.50 | — | 7/7suite scores | |
| 41 | Gemini 3.1 ProGoogle · effort: high | 54.91 | $0.698 | 7/7suite scores | |
| 42 | MiniMax M3MiniMax · effort: max · source-reported open weights | 53.74 | $0.251 | 7/7suite scores | |
| 43 | InklingThinking Machines · effort: max | 51.91 | $0.901 | 7/7suite scores | |
| 44 | Nemotron 3.5 Super VLNVIDIA · effort: max · source-reported open weights | 51.47 | $0.140 | 7/7suite scores | |
| 45 | Mistral Large 4Mistral AI · effort: high | 47.70 | $0.673 | 7/7suite scores | |
| 46 | Claude Haiku 4.5Anthropic · effort: max | 46.04 | — | 7/7suite scores | |
| 47 | Gemma 4 31BGoogle · effort: max · source-reported open weights | 41.53 | $0.071 | 7/7suite scores | |
| 48 | Nemotron 3 UltraNVIDIA · effort: max · source-reported open weights | 40.83 | $0.506 | 7/7suite scores | |
| 49 | Nemotron 3.5 LightningNVIDIA · effort: max · source-reported open weights | 28.69 | $0.053 | 7/7suite scores |
Compare the tradeoffs
Same release and suite definitions. A higher score does not establish a better fit for every workflow.
Inspect a model’s evidence
Each panel records the exact source model ID, effort label, suite scores and recorded cost context. Provider and open-weight labels come from the benchmark manifest.
Claude Fable 5.1 · Anthropic · 7/7 suite scores
Within this run: Browser is its highest-scoring suite (100.0), while Data Analysis is lowest (80.3). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 80.8 | $51.573 |
| Reasoning | 91.7 | $0.460 |
| Coding | 97.7 | $5.282 |
| Browser | 100.0 | $3.746 |
| Data Analysis | 80.3 | $1.333 |
| Research | 82.0 | $20.886 |
| Design | 83.4 | $7.376 |
Source model ID: claude-fable-5.1. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Claude Opus 5.5 · Anthropic · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (96.3), while Browser is lowest (80.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 85.0 | $38.009 |
| Reasoning | 92.2 | $0.252 |
| Coding | 96.3 | $5.031 |
| Browser | 80.0 | $14.670 |
| Data Analysis | 80.3 | $0.815 |
| Research | 85.0 | $8.818 |
| Design | 86.9 | $5.748 |
Source model ID: claude-opus-5.5. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
GPT-6 Astra · OpenAI · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (93.7), while Design is lowest (82.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 82.4 | $38.379 |
| Reasoning | 92.7 | $0.390 |
| Coding | 93.7 | $3.315 |
| Browser | 84.0 | $5.910 |
| Data Analysis | 83.0 | $0.699 |
| Research | 85.2 | $17.660 |
| Design | 82.0 | $2.876 |
Source model ID: gpt-6-astra. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
GPT-6.1 Sol · OpenAI · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (96.0), while Long Agentic is lowest (73.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 73.0 | $6.488 |
| Reasoning | 92.6 | $0.080 |
| Coding | 96.0 | $0.482 |
| Browser | 88.0 | $2.437 |
| Data Analysis | 82.7 | $0.145 |
| Research | 83.5 | $2.415 |
| Design | 82.1 | $0.371 |
Source model ID: gpt-6.1-sol. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Gemini 4 Argon · Google · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (95.1), while Long Agentic is lowest (69.9). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 69.9 | $13.842 |
| Reasoning | 95.1 | $0.078 |
| Coding | 89.7 | $5.076 |
| Browser | 92.0 | $3.434 |
| Data Analysis | 85.2 | $0.284 |
| Research | 83.1 | $5.956 |
| Design | 73.7 | $1.160 |
Source model ID: gemini-4-argon. Effort label: high. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Claude Fable 5 · Anthropic · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (98.3), while Long Agentic is lowest (53.2). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 53.2 | $59.737 |
| Reasoning | 89.7 | $0.602 |
| Coding | 98.3 | $8.927 |
| Browser | 92.0 | $10.302 |
| Data Analysis | 80.5 | $1.301 |
| Research | 86.9 | $17.274 |
| Design | 84.7 | $8.549 |
Source model ID: claude-fable-5. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Claude Sonnet 5.5 · Anthropic · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (94.7), while Data Analysis is lowest (59.5). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 78.8 | $35.011 |
| Reasoning | 91.6 | $0.209 |
| Coding | 94.7 | $4.022 |
| Browser | 84.0 | $7.659 |
| Data Analysis | 59.5 | $0.531 |
| Research | 83.6 | $8.186 |
| Design | 85.0 | $5.017 |
Source model ID: claude-sonnet-5.5. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Claude Opus 5 · Anthropic · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (91.2), while Long Agentic is lowest (62.1). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 62.1 | $48.753 |
| Reasoning | 91.2 | $0.168 |
| Coding | 90.8 | $3.922 |
| Browser | 84.0 | $2.965 |
| Data Analysis | 74.6 | $0.746 |
| Research | 88.9 | $16.970 |
| Design | 83.0 | $5.453 |
Source model ID: claude-opus-5. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
GPT-6 Sol · OpenAI · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (93.1), while Long Agentic is lowest (66.9). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 66.9 | $8.621 |
| Reasoning | 88.7 | $0.121 |
| Coding | 93.1 | $0.977 |
| Browser | 92.0 | $3.284 |
| Data Analysis | 81.2 | $0.199 |
| Research | 69.3 | $3.505 |
| Design | 82.2 | $0.425 |
Source model ID: gpt-6-sol. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
DeepSeek V4.1 Flash 1017 · DeepSeek · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (96.0), while Long Agentic is lowest (50.9). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 50.9 | $1.458 |
| Reasoning | 89.1 | $0.024 |
| Coding | 96.0 | $0.196 |
| Browser | 76.0 | $0.245 |
| Data Analysis | 79.5 | $0.033 |
| Research | 79.5 | $0.375 |
| Design | 83.4 | $0.212 |
Source model ID: deepseek-v4.1-flash-1017. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Muse Wonder 1.0 · Meta · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (90.6), while Long Agentic is lowest (53.2). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 53.2 | $14.483 |
| Reasoning | 90.6 | $0.298 |
| Coding | 86.2 | $3.102 |
| Browser | 72.0 | $4.138 |
| Data Analysis | 81.6 | $0.366 |
| Research | 88.0 | $3.506 |
| Design | 75.1 | $0.687 |
Source model ID: muse-wonder-1.0. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
DeepSeek V4.1 Flash · DeepSeek · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (90.2), while Long Agentic is lowest (40.1). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 40.1 | $1.742 |
| Reasoning | 86.7 | $0.046 |
| Coding | 90.2 | $0.238 |
| Browser | 84.0 | $0.349 |
| Data Analysis | 79.3 | $0.071 |
| Research | 86.8 | $0.409 |
| Design | 79.2 | $0.175 |
Source model ID: deepseek-v4.1-flash. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Claude Opus 4.8 · Anthropic · 7/7 suite scores
Within this run: Browser is its highest-scoring suite (96.0), while Long Agentic is lowest (48.3). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 48.3 | $51.970 |
| Reasoning | 89.2 | $0.606 |
| Coding | 77.3 | $4.152 |
| Browser | 96.0 | $8.872 |
| Data Analysis | 66.0 | $0.646 |
| Research | 81.8 | $6.235 |
| Design | 78.8 | $6.450 |
Source model ID: claude-opus-4.8. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
GPT-5.6 Sol · OpenAI · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (91.7), while Long Agentic is lowest (48.5). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 48.5 | $15.962 |
| Reasoning | 91.7 | $0.191 |
| Coding | 89.7 | $3.769 |
| Browser | 88.0 | $1.338 |
| Data Analysis | 79.8 | $0.238 |
| Research | 54.5 | $14.426 |
| Design | 81.6 | $1.706 |
Source model ID: gpt-5.6-sol. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Claude Haiku 5.5 · Anthropic · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (91.4), while Data Analysis is lowest (53.1). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 57.6 | $10.362 |
| Reasoning | 82.4 | $0.021 |
| Coding | 91.4 | $0.874 |
| Browser | 80.0 | $0.460 |
| Data Analysis | 53.1 | $0.031 |
| Research | 81.9 | $1.940 |
| Design | 83.1 | $0.663 |
Source model ID: claude-haiku-5.5. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
GPT-5.6 Terra · OpenAI · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (90.6), while Long Agentic is lowest (48.4). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 48.4 | $11.757 |
| Reasoning | 90.6 | $0.187 |
| Coding | 86.2 | $2.715 |
| Browser | 84.0 | $0.847 |
| Data Analysis | 79.3 | $0.222 |
| Research | 60.4 | $7.719 |
| Design | 76.0 | $0.620 |
Source model ID: gpt-5.6-terra. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Grok 4.6 · xAI · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (91.2), while Long Agentic is lowest (40.7). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 40.7 | $10.626 |
| Reasoning | 91.2 | $0.087 |
| Coding | 89.1 | $1.858 |
| Browser | 64.0 | $2.921 |
| Data Analysis | 75.3 | $0.123 |
| Research | 84.5 | $2.325 |
| Design | 79.8 | $1.267 |
Source model ID: grok-4.6. Effort label: xhigh. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
MiMo V2.6 Pro · Xiaomi · 7/7 suite scores
Within this run: Browser is its highest-scoring suite (88.0), while Long Agentic is lowest (47.5). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 47.5 | $1.389 |
| Reasoning | 79.6 | $0.024 |
| Coding | 81.5 | $0.224 |
| Browser | 88.0 | $0.457 |
| Data Analysis | 72.3 | $0.039 |
| Research | 76.7 | $0.142 |
| Design | 78.5 | $0.093 |
Source model ID: mimo-v2.6-pro. Effort label: high. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Muse Spark 1.3 · Meta · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (90.2), while Long Agentic is lowest (40.4). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 40.4 | $8.839 |
| Reasoning | 89.7 | $0.136 |
| Coding | 90.2 | $1.372 |
| Browser | 80.0 | $1.690 |
| Data Analysis | 79.6 | $0.169 |
| Research | 68.6 | $2.694 |
| Design | 74.8 | $0.829 |
Source model ID: muse-spark-1.3. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Grok 4.7 · xAI · 7/7 suite scores
Within this run: Research is its highest-scoring suite (86.4), while Long Agentic is lowest (50.1). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 50.1 | $18.933 |
| Reasoning | 82.7 | $0.385 |
| Coding | 84.5 | $4.880 |
| Browser | 60.0 | $4.778 |
| Data Analysis | 76.9 | $0.488 |
| Research | 86.4 | $4.577 |
| Design | 78.0 | $3.370 |
Source model ID: grok-4.7. Effort label: xhigh. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
MiniMax M3.1 Flash · MiniMax · 7/7 suite scores
Within this run: Browser is its highest-scoring suite (92.0), while Long Agentic is lowest (34.7). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 34.7 | $1.280 |
| Reasoning | 82.9 | $0.011 |
| Coding | 83.9 | $0.246 |
| Browser | 92.0 | $0.227 |
| Data Analysis | 71.4 | $0.012 |
| Research | 72.7 | $0.266 |
| Design | 74.0 | $0.158 |
Source model ID: minimax-m3.1-flash. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Claude Sonnet 5 · Anthropic · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (88.7), while Long Agentic is lowest (39.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 39.0 | $33.211 |
| Reasoning | 88.7 | $0.255 |
| Coding | 74.7 | $3.483 |
| Browser | 72.0 | $6.290 |
| Data Analysis | 71.7 | $0.310 |
| Research | 82.8 | $5.961 |
| Design | 82.0 | $5.472 |
Source model ID: claude-sonnet-5. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
GLM-5.3 Flash · Z.AI · 7/7 suite scores
Within this run: Browser is its highest-scoring suite (84.0), while Long Agentic is lowest (28.4). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 28.4 | $1.629 |
| Reasoning | 77.6 | $0.014 |
| Coding | 80.6 | $0.170 |
| Browser | 84.0 | $0.177 |
| Data Analysis | 76.4 | $0.024 |
| Research | 70.1 | $0.187 |
| Design | 77.2 | $0.063 |
Source model ID: glm-5.3-flash. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Grok 4.5 · xAI · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (94.3), while Long Agentic is lowest (27.9). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 27.9 | $4.498 |
| Reasoning | 87.2 | $0.088 |
| Coding | 94.3 | $0.959 |
| Browser | 44.0 | $2.918 |
| Data Analysis | 73.0 | $0.114 |
| Research | 85.9 | $0.818 |
| Design | 78.1 | $0.388 |
Source model ID: grok-4.5. Effort label: xhigh. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
GPT-6 Luna · OpenAI · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (84.5), while Research is lowest (43.5). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 48.1 | $0.968 |
| Reasoning | 81.8 | $0.021 |
| Coding | 84.5 | $0.095 |
| Browser | 72.0 | $0.238 |
| Data Analysis | 73.4 | $0.017 |
| Research | 43.5 | $0.355 |
| Design | 82.5 | $0.031 |
Source model ID: gpt-6-luna. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Qwen3.8 Flash · Alibaba · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (87.4), while Long Agentic is lowest (37.1). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 37.1 | $1.872 |
| Reasoning | 87.4 | $0.022 |
| Coding | 87.4 | $0.106 |
| Browser | 48.0 | $0.224 |
| Data Analysis | 74.2 | $0.043 |
| Research | 74.6 | $0.317 |
| Design | 72.8 | $0.128 |
Source model ID: qwen3.8-flash. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
GPT-5.5 · OpenAI · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (89.7), while Long Agentic is lowest (44.6). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 44.6 | $13.851 |
| Reasoning | 89.7 | $0.289 |
| Coding | 85.1 | $4.208 |
| Browser | 48.0 | $5.620 |
| Data Analysis | 81.6 | $0.406 |
| Research | 61.6 | $17.138 |
| Design | 70.3 | $1.190 |
Source model ID: gpt-5.5. Effort label: xhigh. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Gemini 3.8 Flash · Google · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (89.3), while Long Agentic is lowest (30.1). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 30.1 | $7.039 |
| Reasoning | 89.3 | $0.148 |
| Coding | 82.2 | $2.296 |
| Browser | 64.0 | $1.213 |
| Data Analysis | 54.0 | $0.238 |
| Research | 79.1 | $2.638 |
| Design | 72.7 | $0.980 |
Source model ID: gemini-3.8-flash. Effort label: high. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Gemini 3.7 Flash · Google · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (87.8), while Long Agentic is lowest (28.3). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 28.3 | $5.558 |
| Reasoning | 87.8 | $0.069 |
| Coding | 80.5 | $1.732 |
| Browser | 56.0 | $1.069 |
| Data Analysis | 68.0 | $0.167 |
| Research | 76.5 | $2.684 |
| Design | 73.2 | $0.623 |
Source model ID: gemini-3.7-flash. Effort label: high. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
GPT-5.6 Luna · OpenAI · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (85.6), while Long Agentic is lowest (30.4). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 30.4 | $1.101 |
| Reasoning | 85.6 | $0.028 |
| Coding | 73.6 | $1.823 |
| Browser | 68.0 | $0.097 |
| Data Analysis | 78.0 | $0.027 |
| Research | 51.0 | $0.821 |
| Design | 76.0 | $0.090 |
Source model ID: gpt-5.6-luna. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Claude Sonnet 4.6 · Anthropic · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (86.4), while Long Agentic is lowest (23.1). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 23.1 | $20.420 |
| Reasoning | 86.4 | $0.313 |
| Coding | 66.6 | $2.629 |
| Browser | 68.0 | $3.375 |
| Data Analysis | 76.1 | $0.417 |
| Research | 63.6 | $3.902 |
| Design | 71.8 | $2.139 |
Source model ID: claude-sonnet-4.6. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Qwen3.8 Max · Alibaba · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (88.2), while Browser is lowest (32.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 42.3 | $14.930 |
| Reasoning | 88.2 | $0.131 |
| Coding | 60.3 | $1.987 |
| Browser | 32.0 | $1.543 |
| Data Analysis | 78.4 | $0.219 |
| Research | 69.9 | $1.955 |
| Design | 78.6 | $0.596 |
Source model ID: qwen3.8-max. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
DeepSeek V4 Flash Vision · DeepSeek · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (85.4), while Coding is lowest (37.6). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 39.0 | $1.041 |
| Reasoning | 85.4 | $0.085 |
| Coding | 37.6 | $0.040 |
| Browser | 64.0 | $0.432 |
| Data Analysis | 79.5 | $0.079 |
| Research | 69.2 | $0.249 |
| Design | 70.7 | $0.059 |
Source model ID: deepseek-v4-flash-vision. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Kimi K3 · Moonshot AI · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (90.7), while Browser is lowest (16.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 42.8 | $17.990 |
| Reasoning | 90.7 | $0.195 |
| Coding | 59.3 | $2.061 |
| Browser | 16.0 | $0.382 |
| Data Analysis | 78.7 | $0.352 |
| Research | 79.0 | $2.405 |
| Design | 77.2 | $1.001 |
Source model ID: kimi-k3. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Qwen3.8 27B · Alibaba · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (86.8), while Long Agentic is lowest (27.9). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 27.9 | $7.184 |
| Reasoning | 80.0 | $0.075 |
| Coding | 86.8 | $0.690 |
| Browser | 32.0 | $1.290 |
| Data Analysis | 76.6 | $0.089 |
| Research | 59.4 | $1.127 |
| Design | 75.9 | $0.384 |
Source model ID: qwen3.8-27b. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Muse Spark 1.2 · Meta · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (90.0), while Long Agentic is lowest (21.3). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 21.3 | $3.949 |
| Reasoning | 90.0 | $0.099 |
| Coding | 78.2 | $1.466 |
| Browser | 36.0 | $2.495 |
| Data Analysis | 76.5 | $0.109 |
| Research | 67.7 | $1.783 |
| Design | 67.1 | $0.856 |
Source model ID: muse-spark-1.2. Effort label: xhigh. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Step 5 (Preview) · StepFun · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (85.6), while Browser is lowest (24.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 27.5 | $10.376 |
| Reasoning | 83.8 | $0.048 |
| Coding | 85.6 | $1.095 |
| Browser | 24.0 | $2.792 |
| Data Analysis | 63.5 | $0.064 |
| Research | 72.8 | $1.229 |
| Design | 76.9 | $0.746 |
Source model ID: step-5-preview. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
GLM-5.3 · Z.AI · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (85.8), while Browser is lowest (0.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 35.9 | $18.382 |
| Reasoning | 85.8 | $0.265 |
| Coding | 82.8 | $2.138 |
| Browser | 0.0 | Not available |
| Data Analysis | 70.2 | $0.373 |
| Research | 73.2 | $1.635 |
| Design | 77.3 | $0.956 |
Source model ID: glm-5.3. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Muse Spark Open · Meta · 7/7 suite scores
Within this run: Research is its highest-scoring suite (80.3), while Long Agentic is lowest (18.9). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 18.9 | $4.834 |
| Reasoning | 73.7 | $0.043 |
| Coding | 73.6 | $1.253 |
| Browser | 36.0 | $2.643 |
| Data Analysis | 60.2 | $0.045 |
| Research | 80.3 | $0.732 |
| Design | 60.3 | $0.356 |
Source model ID: muse-spark-open. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Beam 1.0 · Reflection AI · 7/7 suite scores
Within this run: Coding is its highest-scoring suite (83.9), while Browser is lowest (0.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 17.3 | $14.057 |
| Reasoning | 80.2 | $0.040 |
| Coding | 83.9 | $1.852 |
| Browser | 0.0 | Not available |
| Data Analysis | 66.4 | $0.075 |
| Research | 72.8 | $2.385 |
| Design | 67.9 | $0.375 |
Source model ID: beam-1.0. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Gemini 3.1 Pro · Google · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (84.0), while Browser is lowest (4.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 23.9 | $8.126 |
| Reasoning | 84.0 | $0.148 |
| Coding | 66.7 | $1.291 |
| Browser | 4.0 | $1.777 |
| Data Analysis | 78.5 | $0.260 |
| Research | 65.7 | $0.770 |
| Design | 61.6 | $0.596 |
Source model ID: gemini-3.1-pro. Effort label: high. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
MiniMax M3 · MiniMax · 7/7 suite scores
Within this run: Data Analysis is its highest-scoring suite (76.2), while Browser is lowest (8.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 12.4 | $4.195 |
| Reasoning | 74.5 | $0.019 |
| Coding | 70.9 | $0.455 |
| Browser | 8.0 | $0.429 |
| Data Analysis | 76.2 | $0.029 |
| Research | 68.0 | $0.252 |
| Design | 66.2 | $0.208 |
Source model ID: minimax-m3. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Inkling · Thinking Machines · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (78.3), while Long Agentic is lowest (14.1). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 14.1 | $12.406 |
| Reasoning | 78.3 | $0.104 |
| Coding | 64.0 | $1.877 |
| Browser | 16.0 | $2.396 |
| Data Analysis | 72.8 | $0.145 |
| Research | 59.9 | $1.668 |
| Design | 58.3 | $0.286 |
Source model ID: inkling. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Nemotron 3.5 Super VL · NVIDIA · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (78.8), while Browser is lowest (4.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 15.8 | $2.048 |
| Reasoning | 78.8 | $0.013 |
| Coding | 65.5 | $0.331 |
| Browser | 4.0 | $0.351 |
| Data Analysis | 64.6 | $0.019 |
| Research | 70.9 | $0.167 |
| Design | 60.7 | $0.056 |
Source model ID: nemotron-super-vl-3.5. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Mistral Large 4 · Mistral AI · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (83.9), while Browser is lowest (8.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 11.8 | $9.556 |
| Reasoning | 83.9 | $0.119 |
| Coding | 60.0 | $0.983 |
| Browser | 8.0 | $0.451 |
| Data Analysis | 76.5 | $0.134 |
| Research | 28.1 | $2.327 |
| Design | 65.6 | $0.389 |
Source model ID: mistral-large-4. Effort label: high. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Claude Haiku 4.5 · Anthropic · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (61.7), while Long Agentic is lowest (11.1). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 11.1 | $2.549 |
| Reasoning | 61.7 | Not available |
| Coding | 48.1 | $0.494 |
| Browser | 40.0 | $1.502 |
| Data Analysis | 59.3 | Not available |
| Research | 42.2 | $0.432 |
| Design | 59.9 | $0.156 |
Source model ID: claude-haiku-4.5. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Gemma 4 31B · Google · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (68.3), while Browser is lowest (4.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 13.3 | $0.989 |
| Reasoning | 68.3 | $0.004 |
| Coding | 49.0 | $0.233 |
| Browser | 4.0 | $0.254 |
| Data Analysis | 53.3 | $0.005 |
| Research | 44.4 | $0.038 |
| Design | 58.4 | $0.015 |
Source model ID: gemma-4-31b. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest · Our model-family context
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Nemotron 3 Ultra · NVIDIA · 7/7 suite scores
Within this run: Reasoning is its highest-scoring suite (74.7), while Browser is lowest (0.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 11.2 | $5.247 |
| Reasoning | 74.7 | $0.138 |
| Coding | 43.6 | $1.013 |
| Browser | 0.0 | $0.569 |
| Data Analysis | 54.5 | $0.183 |
| Research | 43.2 | $1.958 |
| Design | 58.6 | $0.069 |
Source model ID: nemotron-ultra-3. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
Nemotron 3.5 Lightning · NVIDIA · 7/7 suite scores
Within this run: Design is its highest-scoring suite (43.9), while Browser is lowest (0.0). Suite difficulty differs, so this spread is a diagnostic starting point rather than proof of task suitability.
| Suite | Score / 100 | USD / measured task |
|---|---|---|
| Long Agentic | 8.3 | $0.560 |
| Reasoning | 42.9 | $0.022 |
| Coding | 22.6 | $0.085 |
| Browser | 0.0 | $0.120 |
| Data Analysis | 43.1 | $0.014 |
| Research | 40.0 | $0.087 |
| Design | 43.9 | $0.030 |
Source model ID: nemotron-lightning-3.5. Effort label: max. Source release: 2026-10-14. Scores describe this benchmark configuration, not every API endpoint or consumer app carrying a similar name.
Check the original release · Original score CSV · Publisher link from source manifest
For a real decision, test a representative input, check the produced artifact and record correction effort. Use our workflow evaluation method rather than treating a small rank difference as a production guarantee.
What this board helps you decide
Use the overall view for a shortlist, then select the suite closest to your actual work. Coding, browser tasks and research briefs exercise different capabilities. Compare two or three runs and inspect where their scores diverge before choosing a pilot.
Read the measurement boundary
These are published LiveBench results, not tests performed by AI Tools Ratings. Suite scores use a 0–100 scale but are not one universal accuracy percentage. We do not add vendor launch claims or infer scores for missing results.
Recorded benchmark cost is a task-run measurement from this release. It is not a current API token quote or subscription price. Equal-weight quality scores and question-weighted costs answer different questions; read the formulas and limitations before comparing value.
Model-family profiles · Editorial comparisons · Report a source or identity mismatch