Leaderboard / Baidu (ERNIE)
ERNIE 4.5 21B-A3B Thinking
Baidu (ERNIE) · ERNIE 4.5 · released 2025-09-08 · data checked 2026-11-01 Needs verification
ERNIE 4.5 21B-A3B Thinking is a 21.8B-parameter Mixture-of-Experts model (3B active per token) from Baidu (ERNIE) (China), released 2025-09-08 under the Apache-2.0 license. It doesn't have enough reported benchmarks for an index score yet. It needs about 14.6 GB of memory at 4-bit with an 8K-token context.
Benchmark figures circulating online conflict (the official card shows them only as an image), so none are listed yet.
Benchmarks
Each bar shows this model's score. The tick marks the best medium model on that benchmark.
No benchmark results for this release yet. Know of one? Send it.
Specifications
- Lab
- Baidu (ERNIE) (CN · China)
- Architecture
- Mixture-of-Experts
- Total parameters
- 21.8B
- Active parameters
- 3B
- Size class
- Medium
- Context window
- 128K tokens
- Input modalities
- text
- Reasoning
- Reasoning (thinking)
- License
- Apache-2.0
- Openness
- Open weights
- Memory to run
- ~14.6 GB at 4-bit, 8K context (more sizes)
- Hugging Face
- 58.4K downloads (30 days) · 794 likes
Compare with
- ERNIE 4.5 21B-A3B Thinking vs EuroLLM 22B Instruct (22.6B, index 19.2)
- ERNIE 4.5 21B-A3B Thinking vs gpt-oss-20b (20.9B, index 64.0)
- ERNIE 4.5 21B-A3B Thinking vs Tri-21B Think (20.7B, index 43.6)
Sources
Something wrong here? Report it.
Run it locally
Uses unsloth/ERNIE-4.5-21B-A3B-Thinking-GGUF (Q4_K_M). Swap the tag for another size: BF16, IQ4_NL, IQ4_XS, Q2_K, Q2_K_L, Q3_K_M, Q3_K_S, Q4_0….
ollama run hf.co/unsloth/ERNIE-4.5-21B-A3B-Thinking-GGUF:Q4_K_Mllama-server -hf unsloth/ERNIE-4.5-21B-A3B-Thinking-GGUF:Q4_K_Mvllm serve baidu/ERNIE-4.5-21B-A3B-ThinkingMemory needed: weights, context cache and runtime buffers.
| Quantization | 4K context | 32K context | 128K context |
|---|---|---|---|
| 16-bit F16 | 45 GB | 46 GB | 52 GB |
| 8-bit Q8_0 | 24 GB | 26 GB | 31 GB |
| 4-bit Q4_K_M | 14 GB | 16 GB | 21 GB |