BenchLeader
Menu
Choose a modelLeaderboardsModelsBenchmarksProvidersReleases16

Terminal-Bench-Science 0.1 (AA)

Scientific computing tasks at a terminal: set up the environment, run the analysis, get the answer out. Run by Artificial Analysis.

As of 18 Oct 2026, GPT-6 Astra leads Terminal-Bench-Science 0.1 (AA) on BenchLeader with 63.3%, ahead of Claude Opus 5.5 at 61.9%, across 32 model configurations with a published result.

Published by
Artificial Analysis
Category
Agents & tools
Index weight
Reference only
Models
32
Data as of
18 Oct 2026

Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.

Top 15
  1. 1GPT-6 Astra (max)63.3%
  2. 2Claude Opus 5.5 (xhigh)61.9%
  3. 3Claude Opus 5.5 (max)59.0%
  4. 4GPT-6.1 Sol (max)58.1%
  5. 5Claude Sonnet 5.5 (max)53.3%
  6. 6Gemini 4 Argon (high)52.9%
  7. 7Claude Sonnet 5.5 (xhigh)52.4%
  8. 8Claude Opus 5.5 (high)49.0%
  9. 9Claude Fable 5.1 (max)43.3%
  10. 10Claude Opus 5.5 (medium)43.3%
  11. 11Claude Sonnet 5.5 (high)34.3%
  12. 12GPT-6 Sol (max)30.0%
  13. 13Claude Opus 5 (max)28.6%
  14. 14Claude Haiku 5.5 (max)20.0%
  15. 15Claude Haiku 5.5 (xhigh)18.1%
32 of 32
#
1GPT-6 Astramax63.3%69.92026-09-03
2Claude Opus 5.5xhigh61.9%69.62026-09-22
3Claude Opus 5.5max59.0%71.92026-09-22
4GPT-6.1 Solmax58.1%69.32026-09-29
5Claude Sonnet 5.5max53.3%67.52026-09-28
6Gemini 4 Argonhigh52.9%71.42026-09-30
7Claude Sonnet 5.5xhigh52.4%66.92026-09-28
8Claude Opus 5.5high49.0%69.72026-09-22
9Claude Fable 5.1max43.3%70.12026-09-01
10Claude Opus 5.5medium43.3%66.82026-09-22
11Claude Sonnet 5.5high34.3%66.12026-09-28
12GPT-6 Solmax30.0%65.82026-09-22
13Claude Opus 5max28.6%68.12026-07-24
14Claude Haiku 5.5max20.0%57.92026-10-07
15Claude Haiku 5.5xhigh18.1%59.92026-10-07
16Qwen3.8 Maxmax11.9%63.92026-09-02
17Muse Spark 1.3max10.9%67.42026-09-02
18Gemini 3.8 Flashhigh10.0%63.52026-09-02
19Claude Haiku 5.5high10.0%59.52026-10-07
20GLM 5.3max9.5%63.82026-08-18
21DeepSeek V4.1 Flashmax9.1%61.72026-09-10
22GPT-6 Lunamax8.6%58.72026-09-22
23Grok 4.6high6.2%63.22026-08-12
24MiMo-V2.6-Pro5.7%65.62026-09-21
25DeepSeek V4 Promax5.7%61.02026-08-13
26GLM 5.3 Flash4.8%62.42026-08-26
27Grok 4.7xhigh4.8%60.72026-09-21
28Step 5 Preview2.4%62.42026-09-18
29Claude Haiku 5.5low1.9%55.02026-10-07
30Qwen3.8 27Bxhigh0.5%56.52026-08-14
31MiniMax M30.5%56.42026-06-01
32K2 Horizon 375B A23B0.0%56.62026-09-03

More agents & tools benchmarks · All LLM benchmarks

Cite as: BenchLeader, “Terminal-Bench-Science 0.1 (AA) leaderboard”, https://www.benchleader.com/benchmarks/aa_terminalbench_science, data as of 18 Oct 2026.

Terminal-Bench-Science 0.1 (AA): questions

What does Terminal-Bench-Science 0.1 (AA) measure?
Scientific computing tasks at a terminal: set up the environment, run the analysis, get the answer out. Run by Artificial Analysis. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads Terminal-Bench-Science 0.1 (AA)?
GPT-6 Astra leads Terminal-Bench-Science 0.1 (AA) with 63.3% as of 18 Oct 2026, ahead of Claude Opus 5.5 at 61.9%.
How many models have Terminal-Bench-Science 0.1 (AA) results?
32 model configurations have a Terminal-Bench-Science 0.1 (AA) result on BenchLeader, all taken from Artificial Analysis.
Who runs Terminal-Bench-Science 0.1 (AA) and how often is it updated?
Terminal-Bench-Science 0.1 (AA) is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
Does Terminal-Bench-Science 0.1 (AA) count toward the BenchLeader Index?
No. Terminal-Bench-Science 0.1 (AA) is shown for reference but left out of the composite index.