Compare AI models
| Specifications | ||
ParametersThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. | 236B | 671B |
Context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider. | 128k | 128k |
| API pricingUSD per 1M tokens · lower wins | ||
Cheapest inputLowest input rate across every provider on OpenRouter, the lab included. The cheapest endpoint may run a quantised build or a shorter context — see "Available from" on the model page. | — | $0.27SiliconFlow |
Cheapest outputLowest output rate across every provider on OpenRouter, the lab included. May come from a different provider than the cheapest input. | — | $0.41SiliconFlow |
| — | $74.50SiliconFlow · fp8 | |
| These models have no shared benchmark scores. | ||
SWE-Bench VerifiedCoding — Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better. | — | 67.8% |
SWE-Bench MultilingualMultilingual coding — Like SWE-Bench, but the coding problems span many programming languages, not just one. Tests how broadly the AI can code. Higher is better. | — | 57.9% |
LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better. | — | 74.1% |
HumanEvalFunction synthesis — 164 small Python problems: the AI is given a function's description and has to write the working function. This was the coding benchmark of the GPT-3.5 and GPT-4 era, before the field moved to fixing real bugs in real repositories. Higher is better. | 84.76% | — |
BrowseCompWeb browsing — Can the AI browse the web and track down hard-to-find answers? Higher is better. | — | 40.1% |
MATHReasoning — accuracy on the original competition mathematics test set | 71.02% | — |
AIME 2025Reasoning — single-answer accuracy on the 2025 invitational mathematics exam without tools | — | 89.3% |
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. | — | 79.9% |
MMLU-ProKnowledge — accuracy on the ten-option MMLU-Pro benchmark | — | 85% |
| Overview | ||
| Company | DeepSeek | DeepSeek |
| Release date | Jun 28 2024 | Sep 29 2025 |
| Access | Open Weight | Open Weight |
| Model details | View model | View model |
Other comparisons
DeepSeek V2-0628vsClaude Fable 5.2DeepSeek V3.2 ExpvsClaude Fable 5.2DeepSeek V2-0628vsGPT-6.2 AstraDeepSeek V3.2 ExpvsGPT-6.2 AstraDeepSeek V2-0628vsGemini 4 HydrogenDeepSeek V3.2 ExpvsGemini 4 HydrogenDeepSeek V2-0628vsMuse EmberDeepSeek V3.2 ExpvsMuse EmberDeepSeek V2-0628vsGrok 4.7DeepSeek V3.2 ExpvsGrok 4.7DeepSeek V2-0628vsMistral Large 4DeepSeek V3.2 ExpvsMistral Large 4Frequently asked questions
DeepSeek V2-0628 and DeepSeek V3.2 Exp don't publish scores on any of the same benchmarks, so there's no direct head-to-head comparison. DeepSeek V2-0628 shipped 458 days before DeepSeek V3.2 Exp, so benchmark comparisons should account for the intervening progress.
DeepSeek V2-0628 has 236B parameters, while DeepSeek V3.2 Exp has 671B. Context windows are 128k (DeepSeek V2-0628) vs 128k (DeepSeek V3.2 Exp).
Direct benchmark comparisons are unavailable — DeepSeek V2-0628 and DeepSeek V3.2 Exp don't publish scores on any of the same benchmarks.