Skip to main content

Benchmark profile

Terminal-Bench 2.0

A benchmark for agentic software engineering tasks executed in real terminal environments. DeepSeek reports it in the agentic section, while BenchLM also mirrors it in coding for models that publish it as a developer-task signal.

Data verified

Benchmark score on Terminal-Bench 2.0 — July 23, 2026

BenchLM mirrors the published score view for Terminal-Bench 2.0. GPT-5.6 Sol leads the public snapshot at 91.9% , followed by Claude Mythos 5 (88.0%) and GPT-5.6 Terra (87.4%). BenchLM does not use these results to rank models overall.

45 modelsCodingCurrentDisplay onlyUpdated July 23, 2026

Benchmark score table (45 models)

Score
1
GPT-5.6 SolOpenAI · Closed
91.9%
2
Claude Mythos 5Anthropic · Closed
88.0%
3
GPT-5.6 TerraOpenAI · Closed
87.4%
4
GPT-5.6 LunaOpenAI · Closed
84.7%
5
Claude Fable 5Anthropic · Closed
84.3%
6
Grok 4.5xAI · Closed
83.3%
7
Sakana Fugu-UltraSakana AI · Closed
82.1%
8
GPT-5.5OpenAI · Closed
82.0%
9
SWE-1.7Cognition · Closed
81.5%
10
GLM-5.2Z.AI · Open weight
81.0%
11
Claude Sonnet 5Anthropic · Closed
80.4%
12
Sakana FuguSakana AI · Closed
80.2%
13
Muse Spark 1.1Meta · Closed
80.0%
14
Ornith-1.0-397BDeepReinforce AI · Open weight
77.5%
15
Gemini 3.5 FlashGoogle · Closed
76.2%
16
Claude Opus 4.8Anthropic · Closed
74.6%
17
Qwen3.7 PlusAlibaba · Closed
70.3%
18
Laguna S 2.1Poolside · Open weight
70.2%
19
Qwen3.7 MaxAlibaba · Closed
69.7%
20
Claude Opus 4.7 (Adaptive)Anthropic · Closed
69.4%
21
Composer 2.5Cursor · Closed
69.3%
22
MiMo-V2.5-ProXiaomi · Closed
68.4%
23
DeepSeek V4 Pro (Max)DeepSeek · Open weight
67.9%
24
Kimi K2.6Moonshot AI · Open weight
66.7%
25
MiniMax M3MiniMax · Open weight
66.0%
26
MiMo-V2.5Xiaomi · Closed
65.8%
27
Qwen 3.6 Max (preview)Alibaba · Closed
65.4%
28
Ornith-1.0-35BDeepReinforce AI · Open weight
64.2%
29
InklingThinking Machines Lab · Open weight
63.8%
30
DeepSeek V4 Pro (High)DeepSeek · Open weight
63.3%
31
Composer 2Cursor · Closed
61.7%
32
Step 3.7 FlashStepFun · Open weight
59.5%
33
Qwen3.6-27BAlibaba · Open weight
59.3%
34
DeepSeek V4 ProDeepSeek · Open weight
59.1%
35
DeepSeek V4 Flash (Max)DeepSeek · Open weight
56.9%
36
DeepSeek V4 Flash (High)DeepSeek · Open weight
56.6%
37
Nemotron 3 UltraNVIDIA · Open weight
56.4%
38
Hy3 PreviewTencent · Open weight
54.4%
39
Gemini 3.5 Flash-LiteGoogle · Closed
54.0%
40
Qwen3.6-35B-A3BAlibaba · Open weight
51.5%
41
DeepSeek V4 FlashDeepSeek · Open weight
49.1%
42
MAI-Thinking-1Microsoft · Closed
46.0%
43
Laguna M.1Poolside · Closed
45.8%
44
Ornith-1.0-9BDeepReinforce AI · Open weight
43.1%
45
Laguna XS.2Poolside · Open weight
35.7%

The published Terminal-Bench 2.0 snapshot places GPT-5.6 Sol first at 91.9%. The third row is 4.5 points behind. The broader top-10 range is 10.9 points, so the table still separates the published systems.

45 models have been evaluated on Terminal-Bench 2.0. The benchmark falls in the Coding category. This category carries a 20% weight in BenchLM.ai's overall scoring system. Terminal-Bench 2.0 is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About Terminal-Bench 2.0

Year

2026

Tasks

Terminal-based software tasks

Format

Interactive CLI agent evaluation

Difficulty

Professional software engineering

Terminal-Bench 2.0 focuses on realistic CLI and repository workflows rather than toy code generation. BenchLM keeps coding-category copies display-only unless the scoring weights include them.

BenchLM freshness & provenance

Version

Terminal-Bench 2

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does Terminal-Bench 2.0 measure?

A benchmark for agentic software engineering tasks executed in real terminal environments. DeepSeek reports it in the agentic section, while BenchLM also mirrors it in coding for models that publish it as a developer-task signal.

Which model scores highest on Terminal-Bench 2.0?

GPT-5.6 Sol by OpenAI currently leads with a score of 91.9% on Terminal-Bench 2.0.

How many models are evaluated on Terminal-Bench 2.0?

45 AI models have been evaluated on Terminal-Bench 2.0 on BenchLM.

Last updated: July 23, 2026 · BenchLM version Terminal-Bench 2

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.