BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

Terminal-Bench 2.1 (Vals)

Terminal-Bench 2.1 in Vals AI's fixed harness.

As of 24 Oct 2026, Claude Opus 5.5 leads Terminal-Bench 2.1 (Vals) on BenchLeader with 87.6%, ahead of GPT-6 Astra at 87.3%, across 73 model configurations with a published result.

Published by
Vals AI
Category
Agents & tools
Index weight
0.5
Models
73
Data as of
24 Oct 2026
Newest result
28 Sept 2026

Vals AI (vals.ai).

What the test looks like

Terminal-Bench gives an agent a shell in a container and a task such as building a project from source, recovering a corrupted file system, or setting up a service, then checks the end state with tests.

How it is scored

Percent of tasks passing their checks.

What to keep in mind

Terminal-Bench 2.1 run by Vals AI in its own fixed harness. Fixed-harness results are comparable across models but can differ from the best published runs. See also the Terminal-Bench board.

Top 15
  1. 1Claude Opus 5.587.6%
  2. 2GPT-6 Astra (max)87.3%
  3. 3GPT-5.6 Sol (max)85.8%
  4. 4Claude Fable 5.185.0%
  5. 5Claude Opus 584.6%
  6. 6GPT-6 Sol (max)83.2%
  7. 7Claude Sonnet 5.583.2%
  8. 8Gemini 3.8 Flash (high)81.3%
  9. 9Kimi K380.9%
  10. 10Claude Fable 580.5%
  11. 11Muse Spark 1.3 (max)79.0%
  12. 12GPT-5.6 Luna (max)79.0%
  13. 13Grok 4.6 (high)78.3%
  14. 14GPT-5.6 Terra (max)77.5%
  15. 15Gemini 3.7 Flash (high)77.5%
73 of 73
#
1Claude Opus 5.587.6%62.02026-09-28
2GPT-6 Astramax87.3%70.42026-09-28
3GPT-5.6 Solmax85.8%68.02026-09-28
4Claude Fable 5.185.0%66.42026-09-28
5Claude Opus 584.6%65.52026-09-28
6GPT-6 Solmax83.2%66.42026-09-28
7Claude Sonnet 5.583.2%–2026-09-28
8Gemini 3.8 Flashhigh81.3%64.42026-09-28
9Kimi K380.9%59.12026-09-28
10Claude Fable 580.5%67.12026-09-28
11Muse Spark 1.3max79.0%67.12026-09-28
12GPT-5.6 Lunamax79.0%59.92026-09-28
13Grok 4.6high78.3%62.92026-09-28
14GPT-5.6 Terramax77.5%64.42026-09-28
15Gemini 3.7 Flashhigh77.5%63.52026-09-28
16GPT-5.5high76.4%65.32026-09-28
17MiMo-V2.6-Flash76.4%59.72026-09-28
18Claude Sonnet 574.5%56.12026-09-28
19DeepSeek V4.1 Flashhigh74.5%55.62026-09-28
20Gemini 3.5 Flashhigh74.2%63.12026-09-28
21Gemini 3.6 Flashhigh73.8%62.02026-09-28
22Grok 4.7xhigh73.4%61.62026-09-28
23GPT-6 Lunamax73.0%60.02026-09-28
24Muse Spark 1.3xhigh72.3%64.92026-09-28
25Claude Opus 4.871.9%60.12026-09-28
26GLM 5.3max71.5%63.52026-09-28
27Gemini 3.1 Prohigh70.8%60.32026-09-28
28Muse Spark 1.2xhigh69.7%63.22026-09-28
29Claude Opus 4.8 Claude Code69.7%–2026-09-28
30Muse Spark 1.1xhigh69.3%61.62026-09-28
31Claude Opus 4.768.5%63.32026-09-28
32MiMo-V2.6-Pro67.8%64.02026-09-28
33GLM 5.2max67.8%62.32026-09-28
34Grok 4.5high67.8%59.32026-09-28
35Qwen3.8 Maxmax67.4%63.32026-09-28
36Kimi K2.7 Code67.0%55.82026-09-28
37DeepSeek V4 Flash67.0%52.22026-09-28
38GLM 5.3 Flashmax62.9%54.22026-09-28
39Qwen3.7max61.0%60.62026-09-28
40Qwen3.8 27Bxhigh58.4%56.22026-09-28
41Composer 2.558.4%–2026-09-28
42Barium Bbhigh57.7%–2026-09-28
43Claude Sonnet 4.657.3%57.12026-09-28
44MiMo-V2.5-Pro57.3%54.92026-09-28
45GPT 5.5 Codex57.3%–2026-09-28
46GPT 5.5 Factory57.3%–2026-09-28
47GLM 5.156.9%55.92026-09-28
48Inkling Small55.1%54.82026-09-28
49DeepSeek V4 Promax54.7%60.82026-09-28
50GPT-5.4 Minihigh54.7%52.22026-09-28
51Gemini 3 Flashhigh53.9%58.32026-09-28
52MiniMax M353.6%56.42026-09-28
53Kimi K2.653.6%55.42026-09-28
54Qwen3.6 Plus53.2%58.82026-09-28
55Qwen3.7 Plus52.8%58.52026-09-28
56Nemotron 3 Ultra 550B A55B50.9%55.52026-09-28
57DeepSeek V4 Pro50.2%53.52026-09-28
58Ling 3.0 Flash50.2%52.82026-09-28
59Gemini 3.5 Flash Litehigh50.2%47.92026-09-28
60Ling 3.0 Flash Af Rc350.2%–2026-09-28

More agents & tools benchmarks · All LLM benchmarks

Cite as: BenchLeader, “Terminal-Bench 2.1 (Vals) leaderboard”, data as of 24 Oct 2026, https://www.benchleader.com/benchmarks/vals_terminal_bench.

Terminal-Bench 2.1 (Vals): questions

What does Terminal-Bench 2.1 (Vals) measure?
Terminal-Bench gives an agent a shell in a container and a task such as building a project from source, recovering a corrupted file system, or setting up a service, then checks the end state with tests. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads Terminal-Bench 2.1 (Vals)?
Claude Opus 5.5 leads Terminal-Bench 2.1 (Vals) with 87.6% as of 24 Oct 2026, ahead of GPT-6 Astra at 87.3%.
How many models have Terminal-Bench 2.1 (Vals) results?
73 model configurations have a Terminal-Bench 2.1 (Vals) result on BenchLeader, all taken from Vals AI.
Who runs Terminal-Bench 2.1 (Vals) and how often is it updated?
Terminal-Bench 2.1 (Vals) is published by Vals AI. BenchLeader re-reads the published results once a day and records the date each result was published.
Does Terminal-Bench 2.1 (Vals) count toward the BenchLeader Index?
Yes. Terminal-Bench 2.1 (Vals) contributes to the agents & tools category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.