InternLM3-8B-Instruct.
Shanghai AI Lab Released January 13th, 2025 Family InternLM3 The lab's model card Compare with others Shanghai AI Lab Jan 2025 Open weights Small
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 | |
|---|---|---|---|---|
| HellaSwag | 91.2% | #6 of 53 | Lab model card | |
| Vectara Hallucination Leaderboard (HHEM) | 4.00% | #66 of 148 | Official leaderboard | |
| DROP | 83.1% | #22 of 41 | Lab model card | |
| AlpacaEval 2.0 | 51.1% | #9 of 16 | Lab model card | |
| RULER | 87.9% | #24 of 42 | Lab model card | |
| MMLU | 76.6% | #91 of 131 | Lab model card | |
| CMMLU | 83.1% | #12 of 17 | Lab model card | |
| IFEval | 79.3% | #107 of 135 | Lab model card | |
| MMLU-Pro | 57.6% | #148 of 167 | Lab model card | |
| WildBench | 33.1 | #4 of 4 | Lab model card | |
| MT-Bench-101 | 8.59 | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveMathBench | 33.6% | #4 of 5 | Official leaderboard | |
| MATH-500 | 83% | #58 of 70 | Lab model card | |
| GPQA Diamond | 37.4% | #279 of 302 | Lab model card | |
| PHYSICS | 4.80% | #19 of 20 | Official leaderboard | |
| AIME 2024 | 20% | #85 of 88 | Lab model card | |
| KOR-Bench | 56.4% | #4 of 4 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HumanEval | 82.3% | #50 of 101 | Lab model card | |
| LiveCodeBench | 17.8% | #50 of 54 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| FinEval 6.0 | 78.5% | #3 of 4 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| VRBench | 47.8% | #17 of 19 | Official leaderboard |
The Self-Reported Index fits one score to the 13 capability results Shanghai AI Lab 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 115.7 to 120.1. How we compute it.
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
The benchmarks the lab chose to headline, with the scores it gave.