Best LLM for olympiad programming (IOI)
- Gemini 4 Argon leads
- 100.0%
- models ranked
- 42
IOI (Vals AI) · read October 21
# IOI (Vals AI), read October 21
leaders = {
"Gemini 4 Argon": 100.0%, "GPT-6 Astra": 100.0%, "GPT-6.1 Sol": 96.9%,
}
As of October 21, 2026, Google Gemini 4 Argon leads 42 models on IOI (Vals AI) with 100.0%, ahead of OpenAI GPT-6 Astra (100.0%) and OpenAI GPT-6.1 Sol (96.9%).
The full best LLM for coding ranking combines this benchmark with others.
| Rank | Model | IOI (Vals AI) | Coding evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Gemini 4 Argon Google | 100.0% | 81.0 | $10.00 | |
| 2 | GPT-6 Astra OpenAI | 100.0% | 85.1 | $50.00 | |
| 3 | GPT-6.1 Sol OpenAI | 96.9% | 82.4 | $10.00 | |
| 4 | Claude Opus 5.5 Anthropic | 95.1% | 88.8 | $20.00 | |
| 5 | GPT-5.6 Sol OpenAI | 91.2% | 73.5 | $20.00 | |
| 6 | Claude Fable 5.1 Anthropic | 90.8% | 84.2 | $50.00 | |
| 7 | GPT-5.6 Terra OpenAI | 87.6% | 63.9 | $12.00 | |
| 8 | Claude Opus 5 Anthropic | 84.3% | 83.0 | $25.00 | |
| 9 | Claude Sonnet 5.5 Anthropic | 83.1% | 87.3 | $10.00 | |
| 10 | GPT-6 Sol OpenAI | 82.6% | 76.0 | $10.00 | |
| 11 | Qwen3.8 Max (0902) Qwen | 68.9% | 62.4 | $6.00 | |
| 12 | GLM 5.3 Z.ai | 68.4% | 70.6 | $4.40 | |
| 13 | Gemini 3.7 Flash Google | 67.8% | 58.0 | $3.75 | |
| 14 | Muse Wonder Meta | 66.9% | 71.9 | $20.00 | |
| 15 | GPT-5.6 Luna OpenAI | 61.8% | 58.0 | $1.20 | |
| 16 | Hy4 preview Tencent | 59.3% | 44.3 | $2.2509 | |
| 17 | Grok 4.7 xAI | 57.7% | 66.7 | $6.00 | |
| 18 | Gemini 3.8 Flash Google | 56.9% | 61.6 | $3.75 | |
| 19 | GPT-6 Luna OpenAI | 55.6% | 57.3 | $0.50 | |
| 20 | GPT-5.3-Codex OpenAI | 53.8% | 60.0 | $14.00 |
How this list is built
Generated from the dataset on every refresh. Nobody picks the order by hand.
Models by their IOI (Vals AI) result, as the board publishes it, best first. Problems from the International Olympiad in Informatics, run by Vals AI. Share of points.
- OpenRouter, fetched
- Vals AI boards, fetched