benchgap
Calibration

Terminal-Bench 2.1 → AA Terminal-Bench 2.1

AA Terminal-Bench 2.1 is estimated from Terminal-Bench 2.1 with a Michaelis–Menten curve fitted on 13 models measured on both: y = 2.0000·x / (1.18458 + x), R² = 0.62, cross-validated error 6.8 pp. It is used for 41 estimates.

Estimated modelTerminal-Bench 2.1AA Terminal-Bench 2.1Source
SWE-292.8%87.9%estimated ± 6.8 pp, low confidence
Claude Mythos 588.0%85.2%estimated ± 6.8 pp, medium confidence
MiMo-V2.6-Flash87.6%85.0%estimated ± 6.8 pp, medium confidence
Ultramarine 1 Flash83.2%82.5%estimated ± 6.8 pp, medium confidence
Sakana Fugu-Ultra82.1%81.9%estimated ± 6.8 pp, medium confidence
Ember-182.0%81.8%estimated ± 6.8 pp, medium confidence
SWE-1.781.5%81.5%estimated ± 6.8 pp, medium confidence
Ling 3.1 Flash81.2%81.3%estimated ± 6.8 pp, medium confidence
Sakana Fugu80.2%80.7%estimated ± 6.8 pp, medium confidence
Particle-179.9%80.6%estimated ± 6.8 pp, medium confidence
Beam 1.0 501B-A23B FP878.9%80.0%estimated ± 6.8 pp, medium confidence
Beam 1.0 501B-A23B NVFP478.4%79.7%estimated ± 6.8 pp, medium confidence
Atria Dawn Preview78.3%79.6%estimated ± 6.8 pp, medium confidence
Ornith-1.0-397B77.5%79.1%estimated ± 6.8 pp, medium confidence
dots3-note Preview75.1%77.6%estimated ± 6.8 pp, medium confidence
Jais-3-375B-A23B73.8%76.8%estimated ± 6.8 pp, medium confidence
OrcaCyber-Nano-1.072.5%75.9%estimated ± 6.8 pp, medium confidence
Seed 2.1 Pro71.0%75.0%estimated ± 6.8 pp, medium confidence
Apodex 1.170.8%74.8%estimated ± 6.8 pp, medium confidence
Laguna S 2.170.2%74.4%estimated ± 6.8 pp, medium confidence
Quasar 438B69.3%73.8%estimated ± 6.8 pp, medium confidence
Seed 2.1 Turbo67.6%72.7%estimated ± 6.8 pp, medium confidence
Jais-3-MoVA-36B-A4B67.3%72.5%estimated ± 6.8 pp, medium confidence
Nemotron 3.5 Super VL 120B A12B66.7%72.0%estimated ± 6.8 pp, medium confidence
Pokee-Isaac 28B65.1%70.9%estimated ± 6.8 pp, medium confidence
Ornith-1.0-35B64.2%70.3%estimated ± 6.8 pp, medium confidence
Jais-3-32B63.7%69.9%estimated ± 6.8 pp, low confidence
MAI-Code-1.1-Flash62.9%69.4%estimated ± 6.8 pp, low confidence
Nemotron 3.5 Super VL 120B A12B NVFP460.8%67.8%estimated ± 6.8 pp, low confidence
Step 3.7 Flash59.5%66.9%estimated ± 6.8 pp, low confidence
Solar Pro 457.0%65.0%estimated ± 6.8 pp, low confidence
Flint54.9%63.3%estimated ± 6.8 pp, low confidence
Nemotron 3.5 Bolt 74B-A9B BF1653.9%62.6%estimated ± 6.8 pp, low confidence
Ternary Bonsai 2 27B52.8%61.7%estimated ± 6.8 pp, low confidence
Muse Glimmer 30B51.7%60.8%estimated ± 6.8 pp, low confidence
Nemotron 3.5 Bolt 74B-A9B51.0%60.2%estimated ± 6.8 pp, low confidence
Jais-3-7B46.4%56.3%estimated ± 6.8 pp, low confidence
Ornith-1.5-9B46.2%56.1%estimated ± 6.8 pp, low confidence
K-EXAONE 2.043.8%54.0%estimated ± 6.8 pp, low confidence
Ornith-1.0-9B43.1%53.4%estimated ± 6.8 pp, low confidence
Jais-3-3.7B41.2%51.6%estimated ± 6.8 pp, low confidence

Fitted on

ModelTerminal-Bench 2.1AA Terminal-Bench 2.1
Claude Fable 584.3%84.6%
Claude Opus 4.874.6%84.6%
Claude Sonnet 580.4%80.5%
Gemini 3.7 Flash85.8%85.8%
Gemini 3.8 Flash89.4%87.6%
GLM-5.388.2%83.9%
GLM-5.3-Flash84.3%84.3%
GPT-5.6 Luna84.7%80.9%
GPT-5.6 Terra87.4%88.0%
Grok 4.583.3%81.6%
Inkling63.8%55.1%
Kimi K388.3%85.0%
Qwen3.8-27B73.0%79.8%