Beam 1.0 Base.
Open weights Flagship
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BIG-Bench Hard | 90.4% | #5 of 73 | Lab model card | |
| WinoGrande | 86.7% | #7 of 47 | Lab model card | |
| HellaSwag | 90.3% | #8 of 53 | Lab model card | |
| TriviaQA | 86.6% | #8 of 40 | Lab model card | |
| MMLU-Pro | 72.6% | #116 of 167 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MATH | 94.9% | #4 of 99 | Lab model card | |
| GPQA Diamond | 72.4% | #164 of 302 | Lab model card | |
| AIME 2026 | 63.7% | #35 of 38 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MBPP | 87.8% | #5 of 71 | Lab model card | |
| HumanEval | 87.3% | #30 of 101 | Lab model card | |
| LiveCodeBench v6 | 63.4% | #73 of 90 | Lab model card | |
| SWE-bench Verified | 49.2% | #132 of 155 | Lab model card | |
| SWE-bench Multilingual | 32.1% | #72 of 73 | Lab model card | |
| SWE-Bench Pro | 28.4% | #81 of 81 | Lab model card |
The Self-Reported Index fits one score to the 14 capability results Reflection reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 129.7 to 134.3. How we compute it.
Every result in the longer documents the lab published with it. 14 of them aren't in the launch post.