DeepSeekGA
DeepSeek V4 Pro 0813
A 1.6T MoE with a vendor-reported 96.4% on SWE-bench Verified and last on agentic coding — read both numbers.
Context
1.05M
Max output
384K
Input /1M
$0.15
Output /1M
$6.50
Live pricing via OpenRouter
Best for
- Verifiable, well-specified coding tasks at low cost per test
- Long-context work (1M in, up to 384K out)
- Teams that benchmark on their own tasks before trusting a headline
Watch out
96.4% on SWE-bench Verified yet last of seven frontier peers on LiveBench agentic coding — a headline score here does not predict agent behaviour. Prices rose up to 12x on 2026-08-16 and now vary peak/off-peak. Vendor figures await independent proof.
For creators. Cheap bulk generation where output is checkable. Do not hand it long autonomous loops on this evidence.
Benchmarks
| swe bench verified | 96.4 |
| terminal bench 2 1 | 87.9 |
| deepswe v1 1 | 62.7 |
Capabilities
- 1M context, up to 384K output tokens
- Hybrid attention aimed at cutting long-context inference cost
- Pretrained on 32T+ tokens (vendor-stated)
- Peak/off-peak billing since 2026-08-16 16:00 UTC
More from DeepSeek
Sources
- https://api-docs.deepseek.com/quick_start/pricing
- https://openrouter.ai/deepseek/deepseek-v4-pro-0813
- https://artificialanalysis.ai/models/deepseek-v4-pro
- https://www.techtimes.com/articles/324241/20260813/deepseek-v4-pro-0813-goes-ga-benchmark-claims-await-independent-proof.htm
- https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813