Leaderboard / Baidu (ERNIE)
ERNIE 4.5 VL 28B-A3B Thinking
Baidu (ERNIE) · ERNIE 4.5 · released 2025-11-07 · data checked 2026-11-01 Needs verification
ERNIE 4.5 VL 28B-A3B Thinking is a 28B-parameter Mixture-of-Experts model (3B active per token) from Baidu (ERNIE) (China), released 2025-11-07 under the Apache-2.0 license. It doesn't have enough reported benchmarks for an index score yet. It needs about 18.5 GB of memory at 4-bit with an 8K-token context.
Small LM Index
—
Needs 2 core benchmarks (has 0)
Parameters
28B
3B active · MoE
Price per 1M tokens
—
No hosted API on OpenRouter; self-host
Context window
128K
tokens
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
- 28B
- Active parameters
- 3B
- Size class
- Medium
- Context window
- 128K tokens
- Input modalities
- text, image, video
- Reasoning
- Reasoning (thinking)
- License
- Apache-2.0
- Openness
- Open weights
- Memory to run
- ~18.5 GB at 4-bit, 8K context (more sizes)
- Hugging Face
- 2.7K downloads (30 days) · 545 likes
Compare with
- ERNIE 4.5 VL 28B-A3B Thinking vs Gemma 3 27B (27.4B, index 32.7)
- ERNIE 4.5 VL 28B-A3B Thinking vs Ternary Bonsai 2 27B (27.4B, index 74.6)
- ERNIE 4.5 VL 28B-A3B Thinking vs Qwen3.8 27B (27B, index 77.0)
Sources
Something wrong here? Report it.
Run it locally
No GGUF build from a well-known publisher yet.
vLLM (GPU server)
vllm serve baidu/ERNIE-4.5-VL-28B-A3B-ThinkingFull-precision weights, for a server GPU.
Memory needed: weights, context cache and runtime buffers.
| Quantization | 4K context | 32K context | 128K context |
|---|---|---|---|
| 16-bit F16 | 57 GB | 59 GB | 64 GB |
| 8-bit Q8_0 | 31 GB | 33 GB | 38 GB |
| 4-bit Q4_K_M | 18 GB | 20 GB | 25 GB |