AIEvalsReproduce

AI models,
measured.

Benchmarks, model specifications,
and the sources behind the scores.

Browse the catalogue
Featured benchmark

Terminal-Bench 4.0

Terminal task successHigher is better
  1. Kimi K3.1 Pro67.17%
  2. Claude Sonnet 5.563.64%
  3. Claude Opus 5.559.60%
  4. GPT-6 Astra59.09%
  5. Gemini 4 Argon57.07%
aaRead 2026-10-24

Top 5 of 138 in this evaluator’s field. Run settings and full results in the benchmark.

48
models
165
benchmarks
8,346
attributed results
43
source sites

Recent releases

All models
Model / labContextInput / outputCompare
Kimi K3.1 ProMoonshot AI/23 October 20261,048,576 tokens$2.00 / $10.00
Kimi K3.1 Pro HighspeedMoonshot AI/23 October 20261,048,576 tokens$4.00 / $20.00
DeepSeek V4.1 Flash 1017DeepSeek/17 October 20261M tokens$0.30 / $1.20
GLM-LegalZ.ai/12 October 2026Not publishedNot published
Claude Haiku 5.5Anthropic/7 October 20261M tokensfrom $0.10 / $0.50
BeamReflection AI/5 October 20261M tokens (trained)Not published

USD per million tokens. Prompt tiers may apply. Release dates come from model publishers.

Capability and evaluation cost

AA Intelligence Index · measured USD per task

More capability ↑Less spend ←
AA Intelligence Index versus measured cost per taskEach point is one model at a saved evaluator snapshot. Horizontal axis is USD per index task; vertical axis is the AA Intelligence Index. This is not API list price. Focus a point for exact values and source date.0$0.0015$2.1930$4.3745$6.5660$8.75Measured cost per Intelligence Index task · USDClaude Sonnet 5.5 · index 56 · $7.603/task · read 30 September 2026Claude Opus 5.5 · index 58 · $5.982/task · read 30 September 2026Claude Fable 5.1 · index 53 · $7.630/task · read 30 September 2026Claude Opus 5 · index 51 · $5.858/task · read 30 September 2026GPT-6 Astra · index 53 · $3.258/task · read 30 September 2026GPT-6 Sol · index 48 · $1.049/task · read 30 September 2026GPT-6.1 Sol · index 52 · $0.724/task · read 30 September 2026GPT-5.6 Sol · index 47 · $1.988/task · read 30 September 2026GPT-5.6 Luna · index 37 · $0.178/task · read 30 September 2026Gemini 3.8 Flash · index 41 · $1.243/task · read 30 September 2026Gemini 3.1 Pro · index 30 · $0.675/task · read 30 September 2026Grok 4.7 · index 46 · $3.738/task · read 30 September 2026DeepSeek V4.1 Flash · index 39 · $0.265/task · read 30 September 2026DeepSeek V4 Pro · index 36 · $0.674/task · read 30 September 2026Kimi K3 · index 44 · $2.000/task · read 30 September 2026Qwen3.8 Max · index 40 · $2.670/task · read 30 September 2026GLM 5.3 · index 45 · $2.006/task · read 30 September 2026Muse Spark 1.3 Max · index 48 · $1.605/task · read 30 September 2026MiniMax M3 · index 29 · $0.508/task · read 30 September 2026Claude Sonnet 5 · index 38 · $5.091/task · read 30 September 2026MiMo V2.6 Pro · index 46 · $0.133/task · read 30 September 2026Mistral Medium 3.5 · index 14 · $0.502/task · read 30 September 2026Step 5 Preview · index 44 · $0.717/task · read 30 September 2026GPT-6 Luna · index 37 · $0.068/task · read 30 September 2026GPT-5.6 Terra · index 42 · $1.399/task · read 30 September 2026GLM 5.3 Flash · index 42 · $0.253/task · read 30 September 2026Qwen3.8 2.4T A95B · index 40 · $2.156/task · read 30 September 2026Qwen3.8-Flash-Next · index 40 · $0.372/task · read 30 September 2026MiMo-V2.6-Flash · index 38 · $0.062/task · read 30 September 2026Qwen3.8 Max (0902) · index 45 · $5.409/task · read 30 September 2026Claude Fable 5 · index 50 · $8.746/task · read 30 September 2026Grok 4.6 · index 44 · $2.324/task · read 30 September 2026Claude Opus 4.8 · index 42 · $4.081/task · read 30 September 2026Gemini 3.7 Flash · index 39 · $0.925/task · read 30 September 2026Muse Spark 1.2 · index 40 · $0.975/task · read 30 September 2026Grok 4.5 · index 39 · $1.036/task · read 30 September 2026GPT-5.5 · index 38 · $2.634/task · read 30 September 2026Gemini 4 Argon · index 53 · $1.990/task · read 1 October 2026Claude Haiku 5.5 · index 43 · $0.213/task · read 8 October 2026Kimi K3.1 Pro · index 55.09 · $2.115/task · read 24 October 2026
40 models with both measurementsHover, tap or focus a point to inspect its source snapshot.

Artificial Analysis only. Cost includes measured workload token usage; it is not a token list price. Saved snapshots have different read dates. Models missing either measurement are omitted. Method & source ↗