Benchmark profile

Vibe Code Bench v1.1 (Vibe Code Bench)

Vals.ai benchmark for evaluating whether models can build complete web applications from natural language specifications in a production-like development environment.

Data verified July 18, 2026

How BenchLM shows Vibe Code Bench v1.1

BenchLM mirrors the public Vals AI Vibe Code Bench v1.1 leaderboard captured from https://www.vals.ai/benchmarks/vibe-code and updated by Vals on July 18, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

Vibe Code Bench v1.1 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.

73 Vals rows1 task viewsprivate datasetTasks: OverallDisplay only

Vibe Code Bench v1.1 on Vals AI Vals methodology Vals home

Vibe Code score on Vibe Code Bench — July 18, 2026

BenchLM mirrors the published vibe code score view for Vibe Code Bench. Claude Fable 5 leads the public snapshot at 90.35% , followed by Kimi K3 (84.96%) and Claude Opus 4.8 (82.72%). BenchLM does not use these results to rank models overall.

Claude Fable 5

Anthropic

anthropic/claude-fable-5

90.35%

Overall —

Kimi K3

Moonshot AI

kimi/kimi-k3

84.96%

Overall —

Claude Opus 4.8

Anthropic

anthropic/claude-opus-4-8

82.72%

Overall —

73 modelsCodingCurrentDisplay onlyUpdated July 18, 2026

Vibe Code score table (73 models)

Score

Claude Fable 5Anthropic

90.35%

Kimi K3Moonshot AI

84.96%

Claude Opus 4.8Anthropic

82.72%

Claude Sonnet 5Anthropic

81.33%

GPT-5.6 SolOpenAI

80.50%

Claude Opus 4.8 Claude CodeAnthropic

77.48%

GPT-5.6 LunaOpenAI

77.06%

Muse Spark 1 1Meta

72.16%

Claude Opus 4.7Anthropic

71.00%

GPT-5.5OpenAI

69.85%

Grok 4.5SpaceXAI

69.00%

GPT-5.6 TerraOpenAI

67.84%

GPT-5.4OpenAI

67.42%

GPT-5.5 FactoryOpenAI

67.39%

GLM 5.2Zhipu AI

63.96%

GPT-5.3 CodexOpenAI

61.77%

GPT-5.5 CodexOpenAI

58.19%

Claude Opus 4.6Anthropic

57.57%

Claude Sonnet 4.6 Claude CodeAnthropic

55.77%

GPT-5.2OpenAI

53.50%

Claude Opus 4.6 ThinkingAnthropic

53.50%

Claude Sonnet 4.6Anthropic

51.48%

DeepSeek V4 ProDeepSeek

49.93%

Composer 2.5Cursor

49.61%

Gemini 3.5 FlashGoogle

48.68%

GPT-5.4OpenAI

48.47%

GPT-5.4 MiniOpenAI

47.97%

Qwen3.7 MaxAlibaba

47.67%

MiniMax M3MiniMax

47.57%

Kimi K2.7 CodeMoonshot AI

47.21%

Qwen3.7 PlusAlibaba

46.39%

Mimo V2.5Xiaomi

42.17%

GPT-5.2 CodexOpenAI

37.91%

Kimi K2.6Moonshot AI

37.89%

Mimo V2.5 ProXiaomi

34.11%

Gemini 3.1 Pro PreviewGoogle

32.03%

GLM 5.1Zhipu AI

31.46%

GPT-5.4 NanoOpenAI

26.10%

Qwen3.6 PlusAlibaba

25.57%

GPT-5.1OpenAI

24.61%

GLM 5 ThinkingZhipu AI

23.36%

Claude Sonnet 4.5 20250929 ThinkingAnthropic

22.62%

GPT-5.1 Codex MaxOpenAI

22.17%

Claude Opus 4.5 20251101 ThinkingAnthropic

20.63%

Gemini 3 Flash PreviewGoogle

20.20%

GPT-5OpenAI

20.09%

Muse SparkMeta

19.67%

Grok 4.3SpaceXAI

19.40%

InklingThinkingmachines

19.21%

Kimi K2.5 ThinkingMoonshot AI

17.54%

Qwen3.5 Plus ThinkingAlibaba

15.74%

MiniMax M2.5MiniMax

14.85%

Gemini 3 Pro PreviewGoogle

14.30%

GPT-5 MiniOpenAI

14.17%

Grok Build 0.1SpaceXAI

13.35%

GPT-5.1 CodexOpenAI

13.12%

Qwen3.6 27bAlibaba

11.94%

MiniMax M2.7MiniMax

11.93%

Claude Haiku 4.5 20251001 ThinkingAnthropic

11.39%

Laguna M.1Poolside

11.04%

Nemotron 3 Ultra 550b A55bNvidia

7.64%

Swe 1 6 FastDevin

7.22%

Laguna Xs.2Poolside

5.21%

DeepSeek V3p2 ThinkingFireworks AI

5.11%

Grok 4.20 0309 ReasoningSpaceXAI

4.06%

Qwen3 MaxAlibaba

3.51%

GLM 4.6Zhipu AI

3.09%

Mistral Medium 3.5Mistral AI

2.89%

Grok 4.1 Fast ReasoningSpaceXAI

1.20%

Gemini 2.5 ProGoogle

0.40%

Gemini 3.1 Flash Lite PreviewGoogle

0.00%

Grok 4 Fast ReasoningSpaceXAI

0.00%

Mistral Small 2603Mistral AI

0.00%

The published Vibe Code Bench snapshot places Claude Fable 5 first at 90.35%. The third row is 7.63 points behind. The broader top-10 range is 20.51 points, so the table still separates the published systems.

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

About Vibe Code Bench

Year

2026

Tasks

End-to-end web application builds

Format

Full-stack app implementation benchmark

Difficulty

End-to-end software delivery

Vibe Code Bench v1.1 asks models to build full web apps with services such as Supabase, Stripe test mode, email, browsing, and file editing available. The score is overall application pass accuracy across private end-to-end app tasks.

Vibe Code Bench: Evaluating AI Models on End-to-End Web Application Development Public benchmark source

BenchLM freshness & provenance

Version

Vibe Code Bench 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 Vibe Code Bench measure?

Vals.ai benchmark for evaluating whether models can build complete web applications from natural language specifications in a production-like development environment.

Which model leads the published Vibe Code Bench snapshot?

Claude Fable 5 currently leads the published Vibe Code Bench snapshot with 90.35% vibe code score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on Vibe Code Bench?

73 AI models are included in BenchLM's mirrored Vibe Code Bench snapshot, based on the public leaderboard captured on July 18, 2026.

Last updated: July 18, 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.