Benchmark
Gaia2 Noise
Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages.
- Models measured
- 12
- Scores recorded
- 16
- Sources
- 1
- IQ weight
- 1.98
Score distribution
- 17%
- Mean
- 17.3%
- Median
- 9.4%
- Std dev
- 6.2%
- 10th percentile
- 31.2%
- 90th percentile
- 3.8%
- Min
This benchmark has a weight of 1.98 for chat models, measured over 12 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | GPT 5high· Gaia2 (Meta ARE) data | 35.4% |
| 2 | Claude Sonnet 4thinking· Gaia2 (Meta ARE) data | 31.2% |
| 3 | GPT 5low· Gaia2 (Meta ARE) data | 28.3% |
| 4 | Claude Sonnet 4· Gaia2 (Meta ARE) data | 27.7% |
| 5 | Gemini 2.5 Pro· Gaia2 (Meta ARE) data | 20.4% |
| 6 | Kimi K2· Gaia2 (Meta ARE) data | 18.8% |
| 7 | DeepSeek V3.1 Terminus· Gaia2 (Meta ARE) data | 17.5% |
| 8 | DeepSeek V3.1· Gaia2 (Meta ARE) data | 17.3% |
| 9 | Grok 4· Gaia2 (Meta ARE) data | 15.6% |
| 10 | GPT OSS 120Bhigh· Gaia2 (Meta ARE) data | 14.6% |
| 11 | GPT 5minimal· Gaia2 (Meta ARE) data | 13.1% |
| 12 | Qwen3 235B (Web)· Gaia2 (Meta ARE) data | 10.8% |
| 13 | Qwen3 235B (Web)thinking· Gaia2 (Meta ARE) data | 6.9% |
| 14 | Llama 4 Maverick· Gaia2 (Meta ARE) data | 6.2% |
| 15 | GPT 4o· Gaia2 (Meta ARE) data | 4.6% |
| 16 | Llama 3.3 70B Instruct· Gaia2 (Meta ARE) data | 3.8% |