DeepSeek
The open-weight family everyone else gets benchmarked against.
DeepSeek publishes the weights for nearly everything it ships, which is why its models became the price anchor for the whole market — a Flash model here costs a fraction of a frontier model and gets close on reasoning and code. The V4 line is a sparse mixture-of-experts design with a million-token context. This row is the model family; DeepSeek's own free chat app has its own entry in Chats.
At a glance — DeepSeek V3 0324
- Cost
- $0.25 in · $1.00 out per 1M tokensBudget
- Memory
- 164K tokens — about 250 pages
- Reads
- Text
- Can do
- Use tools (agents)
- Structured output
- Thinking mode (not supported)
- Good at
- Coding: better than 21% · Agentic work: better than 4% · Overall: better than 12%
- Reliability
- 99.35% uptime, last 24h
- Released
- Mar 24, 2025
Advanced details
- API model ID
deepseek/deepseek-chat-v3-0324- Max output
- 147K tokens
- Knowledge cutoff
- 2024-07-31
- Tokenizer
- DeepSeek
- Coding index
- 21.2
- Agentic work index
- 0.8
- Overall index
- 9.7
Providers
The same model, hosted by different companies. OpenRouter picks one per request and falls back to the next if it fails.
| Provider | Input /1M | Output /1M | Cached /1M | Max output | Uptime 24h |
|---|---|---|---|---|---|
| DeepInfra · fp4Degraded | $0.24 | $0.90 | $0.135 | 16K | 91.59% |
| SiliconFlow · fp8 | $0.25 | $1.00 | — | 147K | 99.35% |
| GMICloud · fp8Degraded | $0.29 | $1.14 | $0.11 | 147K | 95.30% |
Supported parameters
- frequency_penalty
- logit_bias
- logprobs
- max_tokens
- min_p
- presence_penalty
- repetition_penalty
- response_format
- seed
- stop
- structured_outputs
- temperature
- tool_choice
- tools
- top_k
- top_logprobs
- top_p
Live from OpenRouter, refreshed hourly. Scores from Artificial Analysis.
Key info
- Pricing
- Open source
- Category
- Models
- Best for
- Intermediate
- Made by
- DeepSeek
- Weights
- Open
- Good for
- Reasoning and code at a fraction of frontier prices