BenchLeader

Terminal-Bench 2.0 (Vals) leaderboard

As of 15 Nov 2026, the top score on Terminal-Bench 2.0 (Vals) is 73.2% by GPT-5.5 (OpenAI), ahead of Claude Opus 4.8 at 70.0%, across 65 model configurations with a published result (Vals AI, result dated 4 Jun 2026). Terminal-Bench 2.0: agent tasks in a shell, checked by tests, run by Vals AI. Terminal-Bench 2.0 (Vals) is shown for reference only and does not count toward the BenchLeader Index.

Key facts (as of 15 Nov 2026)

  • GPT-5.5 holds the top Terminal-Bench 2.0 (Vals) score at 73.2% (Vals AI, 4 Jun 2026).
  • Claude Opus 4.8 is second on Terminal-Bench 2.0 (Vals) at 70.0% (Vals AI, 4 Jun 2026).
  • Terminal-Bench 2.0 (Vals): Terminal-Bench 2.0 in Vals AI's fixed harness. Run by Vals AI (Vals AI).
  • 65 model configurations have a published Terminal-Bench 2.0 (Vals) result on BenchLeader (BenchLeader, 15 Nov 2026).
  • Terminal-Bench 2.0 (Vals) does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).

Terminal-Bench 2.0 in Vals AI's fixed harness. Run by Vals AI.

Published by
Vals AI
Category
Agents & tools
Index weight
Reference only
Models
65
Data as of
15 Nov 2026

Vals AI (vals.ai).

What the test looks like

Terminal-Bench 2.0: agent tasks in a shell, checked by tests, run by Vals AI.

How it is scored

Percent of tasks passing, run by Vals AI.

What to keep in mind

Superseded by 2.1, kept for comparison.

65 of 65
#
1GPT-5.5high73.2%63.5$0.7612026-06-04
2Claude Opus 4.8max70.0%61.1$0.8842026-06-04
3Claude Opus 4.7high68.5%60.0$0.9762026-06-04
4Gemini 3.5 Flashhigh67.4%62.5$1.192026-06-04
5Gemini 3.1 Prohigh67.4%60.4$0.5162026-06-04
6GPT-5.3-Codexhigh64.0%–$0.3452026-06-04
7Muse Spark59.5%64.2$0.0702026-06-04
8Claude Sonnet 4.6max59.5%56.1$0.5192026-06-04
9Qwen3.7max59.2%60.0$0.7192026-06-04
10Claude Opus 4.5high58.4%59.7$1.182026-06-04
11GPT-5.4high58.4%58.6$0.5122026-06-04
12Claude Opus 4.6max58.4%57.0$1.142026-06-04
13Kimi K2.657.3%55.1$0.1932026-06-04
14DeepSeek V4 Pro56.2%53.0$0.1802026-06-04
15Gemini 3 Prohigh55.1%60.0$0.4562026-06-04
16GLM 5.153.9%55.4$0.1582026-06-04
17Qwen3.6 Maxmax51.7%62.1$0.3202026-06-04
18Gemini 3 Flashhigh51.7%57.4$0.1692026-06-04
19GPT-5.2high51.7%55.8$0.4702026-06-04
20GLM-5thinking49.4%60.2$0.2472026-06-04
21MiniMax M2.747.2%56.6$0.0742026-06-04
22MiniMax M346.1%56.1$0.2162026-06-04
23Qwen3.6 Plus44.9%59.0$0.1462026-06-04
24GPT-5.1high44.9%57.5$0.4082026-06-04
25GPT-5.4 Minihigh44.9%52.0$0.3052026-06-04
26Qwen3.6 27B44.9%47.9$0.2982026-06-04
27Grok 4.343.5%54.5$0.3732026-06-04
28Qwen3.5 Plusthinking41.6%57.4$0.3222026-06-04
29MiniMax M2.541.6%53.7$0.0852026-06-04
30Claude Sonnet 4.5thinking41.6%52.3$0.7512026-06-04
31Kimi K2.5thinking40.5%59.4$0.1402026-06-04
32Grok 4.20thinking40.5%58.3$0.1542026-06-04
33GPT-5.4 Nanohigh39.9%48.3$0.0752026-06-04
34Gemma 4 31Bhigh39.3%––2026-06-04
35GLM 4.738.2%51.7$0.1842026-06-04
36Claude Haiku 4.5thinking38.2%45.7$0.3282026-06-04
37GPT-5high37.1%58.2$0.4772026-06-04
38MiniMax-M2.137.1%53.8$0.0702026-06-04
39Kimi K2 Thinkingthinking37.1%52.3$0.1162026-06-04
40DeepSeek V3.2high36.0%54.2$0.1662026-06-04
41DeepSeek V3.2no reasoning34.8%52.1$0.0632026-06-04
42Laguna M.131.5%37.3–2026-06-04
43Gemini 2.5 Pro30.3%55.2$0.4722026-06-04
44Mistral Medium 3.5high30.3%–$4.712026-06-04
45Grok 4 Fastthinking29.2%51.6$0.0342026-06-04
46Grok 428.1%58.1$0.5532026-06-04
47GLM-4.628.1%50.2$0.1442026-06-04
48Laguna XS.228.1%36.3–2026-06-04
49GPT-5 minihigh27.0%54.6$0.1172026-06-04
50Kimi K225.8%51.1$1.112026-06-04
51Grok 4.1 Fastthinking24.7%55.3$0.0562026-06-04
52Qwen3 Maxmax24.7%54.2$0.4232026-06-04
53Qwen3.5 Flash24.7%51.8$0.1102026-06-04
54Gemini 3.1 Flash Litehigh24.7%48.4$0.0652026-06-04
55DeepSeek V3.122.5%49.0$0.5022026-06-04
56Gemini 2.5 Flash 09 2025thinking21.4%52.8$0.1442026-06-04
57gpt-oss-120b19.1%46.1$0.1322026-06-04
58Trinity Largethinking18.0%47.0$0.3262026-06-04
59Grok 4.1 Fastno reasoning18.0%43.1$0.0332026-06-04
60Mistral Small 4high16.9%–$0.1062026-06-04

AI agent benchmark leaderboard · All LLM benchmarks

Cite as: BenchLeader, “Terminal-Bench 2.0 (Vals) leaderboard”, data as of 15 Nov 2026, https://www.benchleader.com/benchmarks/vals_terminal_bench_2.

Terminal-Bench 2.0 (Vals): questions

What does Terminal-Bench 2.0 (Vals) measure?
Terminal-Bench 2.0: agent tasks in a shell, checked by tests, run by Vals AI. Scores are reported in percent of tasks solved; higher is better.
What is the top score on Terminal-Bench 2.0 (Vals)?
As of 15 Nov 2026, the highest score on Terminal-Bench 2.0 (Vals) is 73.2% by GPT-5.5 (result dated 4 Jun 2026), ahead of Claude Opus 4.8 at 70.0%. Scores come from Vals AI and are re-read daily.
Which AI model leads Terminal-Bench 2.0 (Vals)?
GPT-5.5 leads Terminal-Bench 2.0 (Vals) with 73.2% as of 15 Nov 2026 (result dated 4 Jun 2026), ahead of Claude Opus 4.8 at 70.0%.
How many models have Terminal-Bench 2.0 (Vals) results?
65 model configurations have a Terminal-Bench 2.0 (Vals) result on BenchLeader, all taken from Vals AI.
Who runs Terminal-Bench 2.0 (Vals) and how often is it updated?
Terminal-Bench 2.0 (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 2.0 (Vals) count toward the BenchLeader Index?
No. Terminal-Bench 2.0 (Vals) is shown for reference but left out of the composite index.