- AIME2025 (Avg@32) 78
- BFCL (Accuracy) 65
- ZebraLogic (Accuracy) 89.8
- MUSR (Accuracy) 86.7
- BBH (Accuracy) 87.8
- HumanEval+ (Pass@1) 90.8
- MBPP (Pass@1) 80.2
- IFEval (Prompt Strict Accuracy) 84.3
- Multi-IF (Accuracy) 63.3
- ChineseSimpleQA (Accuracy) 49.1
- WritingBench (critic-score, maximum 10) 86.5
ERNIE-4.5-21B-A3B-Thinking
ERNIE-4.5-21B-A3B-Thinking ranks first on MuSR.
13
Benchmarks with a score
1
Ranked first, out of five or more models
2
In the top three
Baidu Sep 2025 ERNIE-4.5-21B-A3B-Thinking
Open weights Small
131.4
#267 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- AIME 2025 78% #93 /143
- IFEval 84.3% #88 /135
- Berkeley Function Calling Leaderboard (BFCL) 65% #28 /41
Where it ranks
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.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BIG-Bench Hard | 87.8% | #11 of 73 | Lab model card | |
| MuSR | 86.7% | #1 of 6 | Lab model card | |
| WritingBench | 86.5 | #3 of 14 | Lab model card | |
| IFEval | 84.3% | #88 of 135 | Lab model card | |
| Multi-IF | 63.3% | #21 of 28 | Lab model card | |
| Chinese SimpleQA | 49.1% | #13 of 13 | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ZebraLogic | 89.8% | #5 of 23 | Lab model card | |
| AIME 2025 | 78% | #93 of 143 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MBPP | 80.2% | #22 of 71 | Lab model card | |
| HumanEval+ | 90.8% | #9 of 27 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Berkeley Function Calling Leaderboard (BFCL) | 65% | #28 of 41 | Lab model card | |
| SnakeBench | 18 | #147 of 188 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveSecBench | 23.7 | #31 of 34 | Official leaderboard |
Its lab's numbers put it at 131.4, number 267 of 404
The Self-Reported Index fits one score to the 9 capability results Baidu 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 127.4 to 135.4. How we compute it.
- HumanEval+ 90.8
- ZebraLogic 89.8
- BIG-Bench Hard 87.8
- IFEval 84.3
- MBPP 80.2
- AIME 2025 78
- Berkeley Function Calling Leaderboard (BFCL) 65
- Multi-IF 63.3
- Chinese SimpleQA 49.1
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.