ERNIE-4.5-21B-A3B-Thinking is Baidu's upgraded lightweight MoE model, refined to boost reasoning depth and quality for top-tier performance in logical puzzles, math, science, coding, text generation, and expert-level academic benchmarks.
Model details →Baidu: ERNIE 4.5 21B A3B Thinking vs DeepSeek: R1 0528
Baidu: ERNIE 4.5 21B A3B Thinking and DeepSeek: R1 0528 are evenly matched on 6 axes — pick the one whose provider, latency, or licensing fits your stack.
May 28th update to the [original DeepSeek R1](/deepseek/deepseek-r1) Performance on par with [OpenAI o1](/openai/o1), but open-sourced and with fully open reasoning tokens. It's 671B parameters in size, with 37B active in an inference pass.…
Model details →Side-by-side comparison
| Capability | Baidu: ERNIE 4.5 21B A3B Thinking | DeepSeek: R1 0528 | Winner |
|---|---|---|---|
| Context window Maximum number of input tokens the model can attend to in a single request. | 131K | 164K | 🏆 DeepSeek: R1 0528 |
| Input price (per 1M) Cost per million input tokens billed by the provider. | $0.07 | $0.45 | 🏆 Baidu: ERNIE 4.5 21B A3B Thinking |
| Output price (per 1M) Cost per million output tokens billed by the provider. | $0.28 | $2.15 | 🏆 Baidu: ERNIE 4.5 21B A3B Thinking |
| Tool / function calling First-class support for emitting structured tool calls. | — | Yes | 🏆 DeepSeek: R1 0528 |
| Vision input Accepts image inputs alongside text. | — | — | Tie |
| Reasoning mode Internal chain-of-thought / extended-thinking support. | Yes | Yes | Tie |
Frequently asked questions
Is Baidu: ERNIE 4.5 21B A3B Thinking better than DeepSeek: R1 0528?
Baidu: ERNIE 4.5 21B A3B Thinking and DeepSeek: R1 0528 are evenly matched on 6 axes — pick the one whose provider, latency, or licensing fits your stack.
What's the price difference between Baidu: ERNIE 4.5 21B A3B Thinking and DeepSeek: R1 0528?
Input: $0.07 vs $0.45 per 1M tokens. Output: $0.28 vs $2.15 per 1M tokens.
What context windows do Baidu: ERNIE 4.5 21B A3B Thinking and DeepSeek: R1 0528 support?
Baidu: ERNIE 4.5 21B A3B Thinking supports up to 131K tokens. DeepSeek: R1 0528 supports up to 164K tokens.
Do both Baidu: ERNIE 4.5 21B A3B Thinking and DeepSeek: R1 0528 support tool calling?
Baidu: ERNIE 4.5 21B A3B Thinking: not advertised. DeepSeek: R1 0528: yes.