DeepSeek Prover Models by DeepSeek
Last refreshed 2026-05-22. Next refresh: weekly.
Details
Capabilities
About
DeepSeek Prover is DeepSeek’s open language-model line for formal theorem proving in Lean 4. DeepSeek-Prover-V2 is initialized with cold-start data from a recursive proving pipeline that uses DeepSeek-V3 to break hard problems into subgoals, then pairs informal chain-of-thought with formal Lean proofs before a reinforcement-learning stage that rewards correct proofs. The series ships 7B and 671B sizes (the larger built on DeepSeek-V3-Base) and is aimed at bridging informal math reasoning with machine-checkable formal proofs.
Decision facts
- Best fit
- reasoningmath-heavy prompts
- Capability starting point
- DeepSeek Prover V2 with 160k context and reasoning
- Lowest tracked input
- DeepSeek Prover V2 · $0.560/1M · Fireworks AI
- Closest related family
- DeepSeek OCR
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs reasoning, 160k context, and 671B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| DeepSeek Prover V2 | Use when the workload needs reasoning, 160k context, and 671B parameters. | 2025-04 | reasoning160k context671B parameters | Current |
| DeepSeek Prover | Use when the workload needs 7B parameters. | 2024-01 | 7B parameters | Current |
Release Timeline
2 release groupsSpecifications(2 models)
| Model | Released | Context | Parameters | Reasoning |
|---|---|---|---|---|
| DeepSeek Prover V2 | 2025-04 | 160k | 671B | Yes |
| DeepSeek Prover | 2024-01 | — | 7B | No |
Available From(2 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| DeepSeek Prover V2 | Fireworks AI | $0.56 | $1.68 | Serverless |
| DeepSeek Prover V2 | Novita AI | $0.7 | $2.5 | Serverless |
