DeepSeek V4 Flash 0731
deepseek-flash
Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding
ReasoningTool CallingStructured OutputOpen WeightsMIT
Context Window
1M
Max Output
384K
Temperature
Yes
Open Weights
Yes
Knowledge Cutoff
2025-05
Released
2026-07-31
Last Updated
2026-07-31
License
MIT
Modalities
Input:Text
→Output:Text
Benchmarks
| Benchmark | Score | Metric | Harness |
|---|---|---|---|
| Terminal-Bench | 82.7 | pass@1 | - |
| NL2Repo | 54.2 | resolve rate | DeepSeek Harness minimal mode |
| CyberGym | 76.7 | score | DeepSeek Harness minimal mode |
| DeepSWE | 54.4 | resolve rate | DeepSeek Harness minimal mode |
| Toolathlon-Verified | 70.3 | score | DeepSeek Harness minimal mode |
| Agents' Last Exam | 25.2 | score | DeepSeek Harness minimal mode |
| AutomationBench | 25.1 | success rate | - |
| DSBench-FullStack | 68.7 | score | - |
| DSBench-Hard | 59.6 | score | - |
Available Providers (12)
| Provider | Input /1M | Output /1M | Cache Read /1M | Cache Write /1M | Reasoning | Status |
|---|---|---|---|---|---|---|
 Perplexity Agent | Rp 2.325 | Rp 4.649 | Rp 501 | — | effort | — |
 NanoGPT | Rp 895 | Rp 2.861 | Rp 233 | — | effort | — |
 Kilo Gateway | Rp 7.868 | Rp 23.602 | Rp 501 | — | effort | — |
 Novita AI | Rp 7.868 | Rp 23.602 | Rp 501 | — | toggle, effort | — |
 OpenRouter | Rp 235 | Rp 22.887 | Rp 235 | — | toggle, effort | — |
 IteraCompute | Rp 6.080 | Rp 18.775 | Rp 626 | — | ✓ | — |
 Ambient | Rp 1.431 | Rp 3.219 | Rp 287 | Gratis | toggle, effort | — |
 Vercel AI Gateway | Rp 1.359 | Rp 2.736 | Rp 251 | — | effort | — |
 OrcaRouter | Rp 2.629 | Rp 5.275 | Rp 358 | — | toggle, effort | — |
 Ofox | Rp 5.508 | Rp 16.522 | Rp 176 | — | toggle, effort | — |
 Merge Gateway | Rp 626 | Rp 1.252 | Rp 126 | — | toggle, effort | — |
 TensorX | Rp 4.471 | Rp 5.365 | Rp 1.073 | — | toggle | — |