Best LLM on Vibe Code Bench
- Claude Sonnet 5.5 leads
- 92.4%
- models ranked
- 96
- lead over #2
- +0.5 pts
Vibe Code Bench (Vals AI) · read October 21
# Vibe Code Bench (Vals AI), read October 21
leaders = {
"Claude Sonnet 5.5": 92.4%, "Gemini 4 Argon": 91.9%, "Claude Haiku 5.5": 90.4%,
}
As of October 21, 2026, Anthropic Claude Sonnet 5.5 leads 96 models on Vibe Code Bench (Vals AI) with 92.4%, ahead of Google Gemini 4 Argon (91.9%) and Anthropic Claude Haiku 5.5 (90.4%).
The full best LLM for coding ranking combines this benchmark with others.
| Rank | Model | Vibe Code Bench (Vals AI) | Coding evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Claude Sonnet 5.5 Anthropic | 92.4% | 87.3 | $10.00 | |
| 2 | Gemini 4 Argon Google | 91.9% | 81.0 | $10.00 | |
| 3 | Claude Haiku 5.5 Anthropic | 90.4% | 70.7 | $0.50 | |
| 4 | Claude Fable 5 Anthropic | 90.3% | 79.3 | $50.00 | |
| 5 | Claude Opus 5.5 Anthropic | 90.3% | 88.8 | $20.00 | |
| 6 | Claude Fable 5.1 Anthropic | 90.3% | 84.2 | $50.00 | |
| 7 | GPT-6 Astra OpenAI | 89.6% | 85.1 | $50.00 | |
| 8 | GPT-6.1 Sol OpenAI | 88.9% | 82.4 | $10.00 | |
| 9 | Claude Opus 5 Anthropic | 88.4% | 83.0 | $25.00 | |
| 10 | GPT-6 Sol OpenAI | 87.8% | 76.0 | $10.00 | |
| 11 | DeepSeek V4.1 Flash 1017 DeepSeek | 87.7% | 70.3 | $0.60 | |
| 12 | Grok 4.7 xAI | 86.2% | 66.7 | $6.00 | |
| 13 | MiMo-V2.6-Pro Xiaomi | 85.2% | 65.4 | $0.87 | |
| 14 | Kimi K3 Moonshot AI | 85.0% | 62.3 | $15.00 | |
| 15 | DeepSeek V4.1 Flash DeepSeek | 84.7% | 62.5 | $0.78 | |
| 16 | Ember-1 Fireworks | 83.9% | 52.6 | $15.00 | |
| 17 | Muse Wonder Meta | 83.1% | 71.9 | $20.00 | |
| 18 | Muse Spark 1.3 Meta | 82.9% | 57.4 | $4.25 | |
| 19 | Claude Opus 4.8 Anthropic | 82.7% | 65.0 | $25.00 | |
| 20 | DeepSeek V4 Pro 0813 DeepSeek | 82.3% | 56.1 | $1.98 |
How this list is built
Generated from the dataset on every refresh. Nobody picks the order by hand.
Models by their Vibe Code Bench (Vals AI) result, as the board publishes it, best first. Vals AI's Vibe Code board, run with one shared harness; its page describes the tasks. The score as Vals AI publishes it.
- OpenRouter, fetched
- Vals AI boards, fetched