Enter opens the highlighted result. Esc closes.

Terminal-Bench 2.1 leaderboard

GPT-6 Astra leads
87.4%
models measured
11
lead over #2
+3.6 pts
median
78.4%

Leader on Terminal-Bench 2.1 · read November 5

terminal_bench_2_1.py


        # Terminal-Bench 2.1, read November 5
        leaders = {
            "GPT-6 Astra": 87.4%,    "Claude Fable 5": 83.8%,    "GPT-5.5": 83.2%,
        }

On November 5, 2026, 11 of our catalog's models had a Terminal-Bench 2.1 result, and OpenAI GPT-6 Astra led with 87.4%, ahead of Anthropic Claude Fable 5 at 83.8%.

Terminal-Bench 2.1 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 2.1 leaderboard: the share of tasks the agent completes. Each version of Terminal-Bench is a different task set, so 2.1 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

11 models with a result

Original board: Terminal-Bench (tbench.ai)
Terminal-Bench 2.1 results for models in the catalog
# Model Result Evidence Coding index Output $/Mtok
1 GPT-6 AstraOpenAI 87.4% 85.1 $50.00
2 Claude Fable 5Anthropic 83.8% 75.9 $50.00
3 GPT-5.5OpenAI 83.2% 67.9 $30.00
4 Grok 4.5xAI 79.3% 52.3 $6.00
5 Claude Opus 4.8Anthropic 78.9% 62.8 $25.00
6 GPT-5.6 TerraOpenAI 78.4% 63.8 $12.00
7 Muse Spark 1.1Meta 76.2% 70.3 $4.25
8 GPT-5.6 LunaOpenAI 75.7% 55.3 $1.20
9 Claude Sonnet 5Anthropic 74.6% 54.7 $10.00
10 Claude Opus 4.7Anthropic 68.9% 67.5 $25.00
11 GLM 5.1Z.ai 58.7% 49.2 $4.40

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 2.1 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 2.1 scores reported by model makers
Rank Model Score Setting Reported by Date
1 DeepSeek V4.1 Flash 90.6% Pass@1 DeepSeek September 10, 2026 (model release)
2 MiMo-V2.6-Pro 89.9% Not published Xiaomi September 21, 2026 (model release)
3 GPT-5.6 Sol 88.8% Not published OpenAI August 7, 2025 (model release)
4 Kimi K3 88.3% max Moonshot AI July 16, 2026 (model release)
5 GLM 5.3 88.2% Not published Z.ai August 18, 2026 (model release)
6 MiMo-V2.6-Flash 87.6% Not published Xiaomi September 21, 2026 (model release)
7 GPT-5.6 Terra 87.4% Not published OpenAI August 7, 2025 (model release)
8 Qwen3.8 Max (0902) 86.6% Not published Qwen August 3, 2026
9 GPT-5.5 85.6% Not published OpenAI August 7, 2025 (model release)
10 Hy4 preview 85.4% Not published Tencent August 28, 2026 (model release) · read from the maker's chart
11 Step 5 Preview 85.0% High StepFun October 8, 2026 (model release)
12 GPT-5.6 Luna 84.7% Not published OpenAI August 7, 2025 (model release)
13 GLM 5.3 Flash 84.3% Not published Z.ai August 26, 2026 (model release) · read from the maker's chart
14 DeepSeek V4 Flash Vision Exp 83.9% Not published DeepSeek August 21, 2026 (model release)
15 Grok 4.5 83.3% Not published xAI July 16, 2026
16 Muse Spark 1.2 82.9% Muse Code Meta August 21, 2026 (model release) · read from the maker's chart
17 N Nex-N2.5-Pro 82.7% Not published Nex AGI September 8, 2026 (model release)
18 Ember-1 82.0% Not published Fireworks September 23, 2026
19 GLM 5.2 81.0% Not published Z.ai August 18, 2026 (model release)
20 R Beam 1.0 501B A23B 80.1% BF16 Reflection October 16, 2026 (model release)
21 Muse Spark 1.1 76.2% mini-swe-agent Meta August 21, 2026 (model release) · read from the maker's chart
22 Qwen3.7 Max 74.5% Not published Qwen August 3, 2026
23 N Nex-N2.5-Mini 73.4% Not published Nex AGI September 8, 2026 (model release)
24 DeepSeek V4 Pro 0813 72.1% Preview DeepSeek July 31, 2026 (model release)
25 Hy3 71.7% Not published Tencent July 6, 2026 (model release) · read from the maker's chart
26 LongCat 2.0 70.8% Not published Meituan July 20, 2026 (model release)
27 P Laguna S 2.1Retired 70.2% Not published Poolside July 21, 2026 (model release)
28 MiniMax M3 66.0% Not published MiniMax May 31, 2026 (model release) · read from the maker's chart
29 MiMo-V2.5-Pro 65.2% Not published Xiaomi September 21, 2026 (model release)
30 DeepSeek V4 Flash 0731 61.8% Preview DeepSeek July 31, 2026 (model release)
31 Step 3.7 Flash 59.6% Not published StepFun May 28, 2026 (model release)
32 Hy3 preview 58.0% Not published Tencent July 6, 2026 (model release) · read from the maker's chart
33 Ling 3.0 Flash 57.0% Not published inclusionAI July 23, 2026 (model release) · read from the maker's chart
33 Solar Pro 4 57.0% Not published Upstage August 10, 2026 (model release)
35 Step 3.5 Flash 53.4% Not published StepFun May 28, 2026 (model release)
36 MiniMax M2.7 51.1% Not published MiniMax May 31, 2026 (model release) · read from the maker's chart
37 Nemotron 3 Super 39.6% Not published NVIDIA August 11, 2026 (model release)
38 Ling 3.0 Tiny 27.7% Thinking inclusionAI November 5, 2026 (model release) · read from the maker's chart
39 Nemotron 3.5 Lightning 24.6% Not published NVIDIA August 11, 2026 (model release)
40 Granite 4.2 8B 20.6% Not published IBM Granite August 31, 2026 (model release)

Other reported runs

Terminal-Bench 2.1 scores reported by model makers, other reported runs
Model Score Setting Reported by Date
DeepSeek V4 Flash 0731 82.7% Not published DeepSeek July 31, 2026 (model release)
DeepSeek V4 Pro 0813 87.9% Not published DeepSeek August 12, 2026 (model release)
R Beam 1.0 501B A23B 78.9% FP8 Reflection October 16, 2026 (model release)
R Beam 1.0 501B A23B 78.4% NVFP4 Reflection October 16, 2026 (model release)
Hy3 70.8% Not published Tencent August 28, 2026 (model release) · read from the maker's chart