Benchmark profile

Vals-hosted Terminal-Bench 2.0 mirror (Vals Terminal-Bench 2.0 mirror)

Vals AI hosted Terminal-Bench 2.0 view with easy, medium, and hard task splits.

Data verified June 4, 2026

How BenchLM shows Vals Terminal-Bench 2.0 mirror

BenchLM mirrors the public Vals AI Vals Terminal-Bench 2.0 mirror leaderboard captured from https://www.vals.ai/benchmarks/terminal-bench-2 and updated by Vals on June 4, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

Vals Terminal-Bench 2.0 mirror is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

67 Vals rows4 task viewspublic datasetTasks: Overall, Easy, Medium, HardDisplay only

Vals Terminal-Bench 2.0 mirror on Vals AI Vals methodology Vals home

Vals Terminal-Bench 2.0 mirror score on Vals Terminal-Bench 2.0 mirror — June 4, 2026

BenchLM mirrors the published vals terminal-bench 2.0 mirror score view for Vals Terminal-Bench 2.0 mirror. GPT-5.5 leads the public snapshot at 73.20% , followed by Claude Opus 4.8 (70.04%) and Claude Opus 4.7 (68.54%). BenchLM does not use these results to rank models overall.

GPT-5.5

OpenAI

openai/gpt-5.5

73.20%

Overall —

Claude Opus 4.8

Anthropic

anthropic/claude-opus-4-8

70.04%

Overall —

Claude Opus 4.7

Anthropic

anthropic/claude-opus-4-7

68.54%

Overall —

67 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated June 4, 2026

Vals Terminal-Bench 2.0 mirror score table (67 models)

Score

GPT-5.5OpenAI

73.20%

Claude Opus 4.8Anthropic

70.04%

Claude Opus 4.7Anthropic

68.54%

Gemini 3.5 FlashGoogle

67.42%

Gemini 3.1 Pro PreviewGoogle

67.42%

GPT-5.3 CodexOpenAI

64.05%

Muse SparkMeta

59.55%

Claude Sonnet 4.6Anthropic

59.55%

Qwen3.7 MaxAlibaba

59.18%

Claude Opus 4.5Anthropic

58.43%

Claude Opus 4.6 ThinkingAnthropic

58.43%

GPT-5.4OpenAI

58.43%

Kimi K2.6Moonshot AI

57.30%

DeepSeek V4 ProDeepSeek

56.18%

Gemini 3 Pro PreviewGoogle

55.06%

Claude Opus 4.5 20251101 ThinkingAnthropic

53.93%

GLM 5.1Zhipu AI

53.93%

Gemini 3 Flash PreviewGoogle

51.69%

GPT-5.2OpenAI

51.69%

Qwen3.6 Max PreviewAlibaba

51.69%

GLM 5 ThinkingZhipu AI

49.44%

MiniMax M2.7MiniMax

47.19%

MiniMax M3MiniMax

46.07%

Qwen3.6 PlusAlibaba

44.94%

GPT-5.1OpenAI

44.94%

Qwen3.6 27bAlibaba

44.94%

GPT-5.4 MiniOpenAI

44.94%

Grok 4.3SpaceXAI

43.45%

Claude Sonnet 4.5 20250929 ThinkingAnthropic

41.57%

MiniMax M2.5MiniMax

41.57%

Qwen3.5 Plus ThinkingAlibaba

41.57%

Grok 4.20 0309 ReasoningSpaceXAI

40.45%

Kimi K2.5 ThinkingMoonshot AI

40.45%

GPT-5.4 NanoOpenAI

39.89%

Gemma 4 31b ItGoogle

39.33%

Claude Haiku 4.5 20251001 ThinkingAnthropic

38.20%

GLM 4.7Zhipu AI

38.20%

MiniMax M2.1MiniMax

37.08%

GPT-5OpenAI

37.08%

Kimi K2 ThinkingMoonshot AI

37.08%

DeepSeek V3p2 ThinkingFireworks AI

35.95%

DeepSeek V3p2Fireworks AI

34.83%

Laguna M.1Poolside

31.46%

Gemini 2.5 ProGoogle

30.34%

Mistral Medium 3.5Mistral AI

30.34%

Grok 4 Fast ReasoningSpaceXAI

29.21%

Laguna Xs.2Poolside

28.09%

GLM 4.6Zhipu AI

28.09%

Grok 4 0709SpaceXAI

28.09%

GPT-5 MiniOpenAI

26.97%

Moonshotai Kimi K2 InstructTogether AI

25.84%

Grok 4.1 Fast ReasoningSpaceXAI

24.72%

Gemini 3.1 Flash Lite PreviewGoogle

24.72%

Qwen3.5 FlashAlibaba

24.72%

Qwen3 MaxAlibaba

24.72%

DeepSeek V3p1Fireworks AI

22.47%

Gemini 2.5 Flash Preview 09 2025 ThinkingGoogle

21.35%

Qwen3 MaxAlibaba

20.23%

GPT Oss 120bFireworks AI

19.10%

Grok 4.1 Fast Non ReasoningSpaceXAI

17.98%

Trinity Large ThinkingArcee-Ai

17.98%

Mistral Small 2603Mistral AI

16.85%

GPT-4.1OpenAI

14.61%

Magistral Medium 2509Mistral AI

13.48%

Mistral Large 2512Mistral AI

8.99%

Command A 03 2025Cohere

2.25%

Llama4 Maverick Instruct BasicFireworks AI

2.25%

The published Vals Terminal-Bench 2.0 mirror snapshot places GPT-5.5 first at 73.20%. The third row is 4.66 points behind. The broader top-10 range is 14.77 points, so the table still separates the published systems.

67 models have been evaluated on Vals Terminal-Bench 2.0 mirror. The benchmark falls in the External benchmark mirrors category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Vals Terminal-Bench 2.0 mirror is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About Vals Terminal-Bench 2.0 mirror

Year

2026

Tasks

Terminal task difficulty splits

Format

Accuracy score

Difficulty

Terminal-based agent execution

BenchLM mirrors this Vals-hosted Terminal-Bench view as display-only secondary context.

Vals Terminal-Bench 2.0 Public benchmark source

BenchLM freshness & provenance

Version

Vals Terminal-Bench 2.0 mirror 2026

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 Vals Terminal-Bench 2.0 mirror measure?

Vals AI hosted Terminal-Bench 2.0 view with easy, medium, and hard task splits.

Which model leads the published Vals Terminal-Bench 2.0 mirror snapshot?

GPT-5.5 currently leads the published Vals Terminal-Bench 2.0 mirror snapshot with 73.20% vals terminal-bench 2.0 mirror score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on Vals Terminal-Bench 2.0 mirror?

67 AI models are included in BenchLM's mirrored Vals Terminal-Bench 2.0 mirror snapshot, based on the public leaderboard captured on June 4, 2026.

Last updated: June 4, 2026 · mirrored from the public benchmark leaderboard

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.