Enter opens the highlighted result. Esc closes.

Terminal-Bench 2.0 leaderboard

GPT-5.5 leads
84.7%
models measured
36
lead over #2
+4.5 pts
median
42.9%

Leader on Terminal-Bench 2.0 · read November 5

terminal_bench_2_0.py


        # Terminal-Bench 2.0, read November 5
        leaders = {
            "GPT-5.5": 84.7%,    "Claude Opus 4.7": 80.2%,    "GPT-5.3-Codex": 78.4%,
        }

On November 5, 2026, 36 of our catalog's models had a Terminal-Bench 2.0 result, and OpenAI GPT-5.5 led with 84.7%, ahead of Anthropic Claude Opus 4.7 at 80.2%.

Terminal-Bench 2.0 gives an AI agent tasks to finish in a real command-line terminal. The number is the accuracy the Terminal-Bench team publishes on its official 2.0 leaderboard: the share of tasks the agent completes. Each version of Terminal-Bench is a different task set, so 2.0 results are not comparable with the other versions.

How to read the score: Each entry on the board is a model run inside an agent harness (Claude Code, Codex, mini-SWE-agent and others), at one reasoning effort. We show the best result per model and name the agent and the effort. The best result can come from an agent the board lists from a third party, not the maker.

Role in the index

A coding board we show beside the Vibe Coding Index, not inside it: the index keeps the six boards it was defined on, so a score does not move every time a board is added. Read the methodology

36 models with a result

Original board: Terminal-Bench (tbench.ai)
Terminal-Bench 2.0 results for models in the catalog
# Model Result Evidence Coding index Output $/Mtok
1 GPT-5.5OpenAI 84.7% 67.9 $30.00
2 Claude Opus 4.7Anthropic 80.2% 67.5 $25.00
3 GPT-5.3-CodexOpenAI 78.4% 60.0 $14.00
4 Claude Opus 4.6Anthropic 76.4% 57.2 $25.00
5 GPT-5.2-CodexOpenAI 66.5% 45.4 $14.00
6 GPT-5.2OpenAI 64.9% 55.9 $14.00
7 Claude Opus 4.5Anthropic 63.1% 44.0 $25.00
8 GPT-5.1-Codex-MiniOpenAI 61.6% Not benchmarked $2.00
9 GPT-5.1-Codex-MaxOpenAI 60.4% Not benchmarked $10.00
10 GPT-5.1-CodexOpenAI 57.8% 15.3 $10.00
11 Grok 4.20xAI 57.3% 33.3 $2.50
12 Claude Sonnet 4.6Anthropic 53.4% 55.0 $15.00
13 GLM 5Z.ai 52.4% 41.7 $3.20
14 GPT-5OpenAI 49.6% 38.8 $10.00
15 GPT-5.1OpenAI 47.6% 40.4 $10.00
16 Claude Sonnet 4.5Anthropic 46.5% 39.5 $15.00
17 MiniMax M2.7MiniMax 45.1% 36.2 $1.20
18 Kimi K2.5Moonshot AI 43.2% 38.6 $2.85
19 MiniMax M2.5MiniMax 42.7% 37.2 $1.20
20 DeepSeek V3.2DeepSeek 39.6% 30.0 $0.42
21 Claude Opus 4.1Anthropic 38.0% Not benchmarked $75.00
22 MiniMax M2.1MiniMax 36.6% Not benchmarked $1.20
23 Kimi K2 ThinkingMoonshot AI 35.7% Not benchmarked $2.50
24 Claude Haiku 4.5Anthropic 35.5% 31.4 $5.00
25 GPT-5 MiniOpenAI 34.8% 34.2 $2.00
26 GLM 4.7Z.ai 33.4% 55.1 $2.20
27 Gemini 2.5 ProGoogle 32.6% 19.1 $10.00
28 MiniMax M2MiniMax 30.0% Not benchmarked $1.02
29 Kimi K2 0711Moonshot AI 27.8% Not benchmarked $2.30
30 Qwen3.6 35B A3BQwen 24.6% Not benchmarked $0.95
31 GLM 4.6Z.ai 24.5% 8.8 $2.20
32 GPT-5 NanoOpenAI 21.8% Not benchmarked $0.40
33 gpt-oss-120bOpenAI 18.7% Not benchmarked $0.50
34 Gemini 2.5 FlashGoogle 17.1% Not benchmarked $2.50
35 Qwen3.5-9BQwen 9.2% Not benchmarked $0.15
36 gpt-oss-20bOpenAI 3.4% Not benchmarked $0.14

Only models in our catalog are listed, and each appears once with its best published run. 62 rows were read from the source on November 5, 2026, and rows that match no catalog model are left out.

Results reported by model makers

The board above is Terminal-Bench (tbench.ai)'s own run. Model makers also report Terminal-Bench 2.0 in their release announcements; those numbers are below, with the date and the announcement. They are not independent and do not count toward any index.

Terminal-Bench 2.0 scores reported by model makers
Rank Model Score Setting Reported by Date
1 Kimi K2.5 50.8% Thinking Moonshot AI January 27, 2026 (model release)
2 GLM 4.7 41.0% Not published Z.ai December 22, 2025 (model release)
3 P Laguna XS 2.1Retired 37.5% Not published Poolside July 2, 2026 (model release)
4 GLM 4.6V 24.5% Not published Z.ai December 22, 2025 (model release)