V4 Flash 0731
DeepSeek · released Jul 31, 2026
DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total. This re-post-trained revision is suited for coding, reasoning, and agent workflows....
Specification
- Context window
- 1.05M
- Max output
- 944K
- Knowledge cutoff
- Not stated
- Parameters
- 304.2B
- Licence
- mit
- Serving providers
- 24
- Moderated
- No
- Uptime
- 100.0%
Intelligence
34.3
87th percentile
Coding
69.1
Coding Index
Agentic
26.3
Agentic Index
Output speed
223 t/s
Median across providers
Latency
422ms
Time to first token
Cost per task
$0.05
Estimated
Benchmarks
Where the score comes from
The Intelligence Index is a composite. These are the underlying evaluations this model was actually measured on.
Evaluation scores
Percentage correct · higher is better
- GPQA Diamond90.8%
- AA-LCR (long context)79.7%
- SciCode50.3%
- τ²-bench (Telecom)39.4%
- Humanity's Last Exam38.6%
An evaluation missing from this list was not run for this model — it is not a zero.
View as table
| Evaluation | Score |
|---|---|
| GPQA Diamond | 90.8% |
| AA-LCR (long context) | 79.7% |
| SciCode | 50.3% |
| τ²-bench (Telecom) | 39.4% |
| Humanity's Last Exam | 38.6% |
Against its peers
Intelligence Index · this model highlighted, nearest peers in grey
- V4 Flash 0731DeepSeek34.3
- V4 Flash 0423DeepSeek34.3
- Gemini 3.6 FlashGoogle34.0
- KAT-Coder-Pro V2KwaiPilot33.9
- Qwen3.8 27BQwen33.7
- Muse Spark 1.1Meta33.7
- GLM 5.2Z.ai33.7
- Gemini 3.5 FlashGoogle32.6
Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.
View as table
| Model | Intelligence |
|---|---|
| V4 Flash 0731 | 34.3 |
| V4 Flash 0423 | 34.3 |
| Gemini 3.6 Flash | 34.0 |
| KAT-Coder-Pro V2 | 33.9 |
| Qwen3.8 27B | 33.7 |
| Muse Spark 1.1 | 33.7 |
| GLM 5.2 | 33.7 |
| Gemini 3.5 Flash | 32.6 |
Percentile among all indexed models
Pricing
What it costs to run
List prices per million tokens, plus what one representative task works out to.
List price
- Input / 1M tokens
- $0.016
- Output / 1M tokens
- $0.64
- Cached input / 1M
- $0.016
- Blended 3:1
- $0.172
One task, estimated
$0.05
- Input tokens
- 50,000
- Output tokens
- 80,000
- Profile
- Reasoning
Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).
Serving providers
- OpenInference
- Relace
- Sail Research
- Reka
- DeepInfra
- DigitalOcean
- Inceptron
- Wafer
- BaseTen
- Venice
- CoreWeave
- StreamLake
- Cohere
- Together
- Parasail
- Mancer 2
- SiliconFlow
- GMICloud
- Alibaba
- Novita
- Phala
- Baidu
- AtlasCloud
- Cloudflare
Speed and latency figures are medians across these providers, so a widely-served model reports a blend rather than any single endpoint.
Arena
Head-to-head generation quality
Elo from pairwise judgements, broken out by the kind of thing the model was asked to build.
- Overall Elo
- 1,239
- Win rate
- 46.7%
- Strongest at
- Websites
- Tournaments
- 6,610
Elo by category
Dot position on a 1,175–1,250 scale · Elo has no meaningful zero
- 1,239
- 1,237
- 1,211
- 1,209
- 1,183
- 1,181
Agent categories (full-stack apps, mobile apps) are only scored for models tested in agent harnesses.
View as table
| Category | Elo | Win rate |
|---|---|---|
| Websites | 1239 | 46.7% |
| UI components | 1237 | 46.8% |
| Game development | 1211 | 44.6% |
| 3D scenes | 1209 | 48.7% |
| SVG | 1183 | 45.1% |
| Data visualisation | 1181 | 41.1% |