Calibration
Terminal-Bench 2.1 → Terminal-Bench 2.1 (Vals AI)
Terminal-Bench 2.1 (Vals AI) is estimated from Terminal-Bench 2.1 with a linear curve fitted on 29 models measured on both: y = 0.7724·x + 0.0597, R² = 0.59, cross-validated error 7.4 pp. It is used for 44 estimates.
| Estimated model | Terminal-Bench 2.1 | Terminal-Bench 2.1 (Vals AI) | Source |
|---|---|---|---|
| Claude Mythos 5 | 88.0% | 73.9% | estimated ± 7.4 pp, medium confidence |
| Ultramarine 1 Flash | 83.2% | 70.2% | estimated ± 7.4 pp, medium confidence |
| Sakana Fugu-Ultra | 82.1% | 69.4% | estimated ± 7.4 pp, medium confidence |
| Ember-1 | 82.0% | 69.3% | estimated ± 7.4 pp, medium confidence |
| SWE-1.7 | 81.5% | 68.9% | estimated ± 7.4 pp, medium confidence |
| Sakana Fugu | 80.2% | 67.9% | estimated ± 7.4 pp, medium confidence |
| Beam 1.0 501B-A23B | 80.1% | 67.8% | estimated ± 7.4 pp, medium confidence |
| Particle-1 | 79.9% | 67.7% | estimated ± 7.4 pp, medium confidence |
| Beam 1.0 501B-A23B FP8 | 78.9% | 66.9% | estimated ± 7.4 pp, medium confidence |
| Beam 1.0 501B-A23B NVFP4 | 78.4% | 66.5% | estimated ± 7.4 pp, medium confidence |
| Atria Dawn Preview | 78.3% | 66.4% | estimated ± 7.4 pp, medium confidence |
| Ornith-1.0-397B | 77.5% | 65.8% | estimated ± 7.4 pp, medium confidence |
| dots3-note Preview | 75.1% | 64.0% | estimated ± 7.4 pp, medium confidence |
| Jais-3-375B-A23B | 73.8% | 63.0% | estimated ± 7.4 pp, medium confidence |
| OrcaCyber-Nano-1.0 | 72.5% | 62.0% | estimated ± 7.4 pp, medium confidence |
| Seed 2.1 Pro | 71.0% | 60.8% | estimated ± 7.4 pp, medium confidence |
| Apodex 1.1 | 70.8% | 60.6% | estimated ± 7.4 pp, medium confidence |
| Laguna S 2.1 | 70.2% | 60.2% | estimated ± 7.4 pp, medium confidence |
| Quasar 438B | 69.3% | 59.5% | estimated ± 7.4 pp, medium confidence |
| Seed 2.1 Turbo | 67.6% | 58.2% | estimated ± 7.4 pp, medium confidence |
| Jais-3-MoVA-36B-A4B | 67.3% | 57.9% | estimated ± 7.4 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B | 66.7% | 57.5% | estimated ± 7.4 pp, medium confidence |
| Pokee-Isaac 28B | 65.1% | 56.2% | estimated ± 7.4 pp, medium confidence |
| Ornith-1.0-35B | 64.2% | 55.6% | estimated ± 7.4 pp, medium confidence |
| Jais-3-32B | 63.7% | 55.2% | estimated ± 7.4 pp, medium confidence |
| MAI-Code-1.1-Flash | 62.9% | 54.5% | estimated ± 7.4 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B NVFP4 | 60.8% | 52.9% | estimated ± 7.4 pp, medium confidence |
| Step 3.7 Flash | 59.5% | 51.9% | estimated ± 7.4 pp, medium confidence |
| Solar Pro 4 | 57.0% | 50.0% | estimated ± 7.4 pp, medium confidence |
| Flint | 54.9% | 48.4% | estimated ± 7.4 pp, medium confidence |
| Nemotron 3.5 Bolt 74B-A9B BF16 | 53.9% | 47.6% | estimated ± 7.4 pp, low confidence |
| Ternary Bonsai 2 27B | 52.8% | 46.7% | estimated ± 7.4 pp, low confidence |
| Muse Glimmer 30B | 51.7% | 45.9% | estimated ± 7.4 pp, low confidence |
| Nemotron 3.5 Bolt 74B-A9B | 51.0% | 45.3% | estimated ± 7.4 pp, low confidence |
| Jais-3-7B | 46.4% | 41.8% | estimated ± 7.4 pp, low confidence |
| Ornith-1.5-9B | 46.2% | 41.6% | estimated ± 7.4 pp, low confidence |
| K-EXAONE 2.0 | 43.8% | 39.8% | estimated ± 7.4 pp, low confidence |
| Ornith-1.0-9B | 43.1% | 39.3% | estimated ± 7.4 pp, low confidence |
| Jais-3-3.7B | 41.2% | 37.8% | estimated ± 7.4 pp, low confidence |
| A.X K2 | 36.0% | 33.8% | estimated ± 7.4 pp, low confidence |
| Granite 4.2 30B | 29.2% | 28.5% | estimated ± 7.4 pp, low confidence |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 23.5% | 24.1% | estimated ± 7.4 pp, low confidence |
| Granite 4.2 8B | 20.6% | 21.8% | estimated ± 7.4 pp, low confidence |
| Mellum2.1-12B-A2.5B-Thinking | 17.4% | 19.4% | estimated ± 7.4 pp, low confidence |
Fitted on
| Model | Terminal-Bench 2.1 | Terminal-Bench 2.1 (Vals AI) |
|---|---|---|
| Claude Fable 5 | 84.3% | 80.5% |
| Claude Opus 4.8 | 74.6% | 71.9% |
| Claude Sonnet 5 | 80.4% | 74.5% |
| DeepSeek V4.1 Flash 1017 | 90.6% | 74.5% |
| DeepSeek V4 Flash 0731 | 82.7% | 67.0% |
| DeepSeek V4 Pro 0813 | 87.9% | 54.7% |
| Gemini 3.5 Flash | 76.2% | 74.2% |
| Gemini 3.5 Flash-Lite | 54.0% | 50.2% |
| Gemini 3.7 Flash | 85.8% | 77.5% |
| Gemini 3.8 Flash | 89.4% | 81.3% |
| GLM-5.2 | 81.0% | 67.8% |
| GLM-5.3 | 88.2% | 71.5% |
| GLM-5.3-Flash | 84.3% | 62.9% |
| GPT-5.6 Luna | 84.7% | 79.0% |
| GPT-5.6 Sol | 91.9% | 85.8% |
| GPT-5.6 Terra | 87.4% | 77.5% |
| Grok 4.5 | 83.3% | 67.8% |
| Hy4 preview | 85.4% | 55.1% |
| Inkling | 63.8% | 47.6% |
| Inkling-Small | 64.7% | 55.1% |
| Kimi K3 | 88.3% | 80.9% |
| Ling 3.0 Flash | 57.0% | 50.2% |
| MiniMax M3 | 66.0% | 53.6% |
| Muse Spark 1.1 | 80.0% | 69.3% |
| Muse Spark 1.2 | 82.9% | 69.7% |
| Muse Spark 1.3 | 88.8% | 72.3% |
| Nemotron 3 Ultra | 56.4% | 50.9% |
| Qwen3.8-27B | 73.0% | 58.4% |
| Qwen3.8 Max | 86.6% | 67.4% |