DeepSeekDeepSeek V4 Flash 0731
DeepSeek V4 Flash 0731 is a DeepSeek open-weights reasoning model, released 31 Jul 2026. Its best configuration (high reasoning effort) ranks #164 of 372 on the BenchLeader Index at 56.3 ±3.1, in the upper half. It scores highest in coding (61) and lowest in knowledge (55). At $0.094 per million tokens blended it is among the cheapest fifth of ranked models. Output speed of 37 tokens per second puts it in the slowest quarter, with a first token in 0.8 s. It has been measured at 5 reasoning-effort settings; tables show the best-scoring one. Last measured 10 Sept 2026.
- Blended price
- $0.094/M
- $0.065 in · $0.180 out
- Output speed
- 37 tok/s
- OpenRouter traffic, 7-day median; not yet measured by Artificial Analysis
- First answer
- 0.84 s
- Context
- 1.3M
- Overall index56
- Reasoning61
- Coding61
- Knowledge55
Versions
DeepSeek has shipped 2 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| DeepSeek V4 Flash 0731highthis page | 31 Jul 2026 | 56.3 | #164 |
| DeepSeek V4 Flashhigh | 24 Apr 2026 | 59.1 | #110 |
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | – | – | 37 tok/s | 0.84 s | $0.0001 | Reasoning 33 |
| low | – | – | 37 tok/s | 0.84 s | $0.0001 | Reasoning 57 |
| highbest | 56.3 | #164 | 37 tok/s | 0.84 s | $0.0001 | Coding 61 · Knowledge 55 · Reasoning 61 |
| max | 54.5 | #201 | 37 tok/s | 0.84 s | $0.0001 | Coding 59 · Knowledge 44 · Maths 57 · Reasoning 64 |
| default | 51.1 | #277 | 37 tok/s | 0.84 s | $0.0001 | Agents & tools 55 · Coding 33 · Composite 49 · Maths 61 · Reasoning 60 |
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | no reasoning | low | high | max | default | Source |
|---|---|---|---|---|---|---|
| GPQA Diamond | – | – | – | 91.0%#28 | – | Epoch AI Benchmarking Hub |
| SimpleBench | – | – | – | – | 61.1%#26 | SimpleBench |
| LiveBench Reasoningnot in index | – | – | – | – | 86.6%#24 | LiveBench |
| GPQA Diamond (Vals)not in index | – | – | 89.9%#27 | – | – | Vals AI |
| ARC-AGI-1 | 11.8%#167 | 84.0%#77 | 87.0%#63 | 89.0%#56 | – | ARC Prize |
| ARC-AGI-2 | 2.1%#141 | 46.0%#75 | 56.0%#64 | 61.4%#53 | – | ARC Prize |
Coding
| Benchmark | no reasoning | low | high | max | default | Source |
|---|---|---|---|---|---|---|
| SciCode | – | – | – | 49.9%#62 | – | SciCode |
| WeirdML | – | – | 57.0%#53 | 63.0%#40 | – | WeirdML |
| FrontierCode | – | – | – | – | 18.8%#24 | Cognition |
| LiveBench Codingnot in index | – | – | – | – | 75.0%#39 | LiveBench |
| LiveCodeBench | – | – | 87.3%#17 | – | – | Vals AI |
| SWE-bench (Vals)not in index | – | – | 88.8%#11 | – | – | Vals AI |
Agents & tools
| Benchmark | no reasoning | low | high | max | default | Source |
|---|---|---|---|---|---|---|
| LiveBench Agentic Codingnot in index | – | – | – | – | 46.8%#38 | LiveBench |
| Terminal-Bench 2.1 (Vals) | – | – | – | – | 67.0%#29 | Vals AI |
Maths
| Benchmark | no reasoning | low | high | max | default | Source |
|---|---|---|---|---|---|---|
| FrontierMath Tiers 1–3 | – | – | – | 57.5%#31 | – | Epoch AI Benchmarking Hub |
| FrontierMath Tier 4 | – | – | – | 24.4%#37 | – | Epoch AI Benchmarking Hub |
| OTIS Mock AIME | – | – | – | 94.4%#41 | – | Epoch AI Benchmarking Hub |
| ProofBench | – | – | – | – | 56.0%#13 | Vals AI |
| LiveBench Mathematicsnot in index | – | – | – | – | 86.8%#38 | LiveBench |
Knowledge
| Benchmark | no reasoning | low | high | max | default | Source |
|---|---|---|---|---|---|---|
| SimpleQA Verified | – | – | – | 33.6%#50 | – | Epoch AI Benchmarking Hub |
| LiveBench Data Analysisnot in index | – | – | – | – | 79.3%#8 | LiveBench |
| MMLU-Pro | – | – | 86.2%#46 | – | – | Vals AI |
| LegalBench | – | – | 77.7%#100 | – | – | Vals AI |
| CorpFin | – | – | 61.9%#52 | – | – | Vals AI |
| TaxEval | – | – | 70.7%#85 | – | – | Vals AI |
Instruction following
| Benchmark | no reasoning | low | high | max | default | Source |
|---|---|---|---|---|---|---|
| LiveBench Languagenot in index | – | – | – | – | 79.2%#29 | LiveBench |
Composite
| Benchmark | no reasoning | low | high | max | default | Source |
|---|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | – | – | 154.5#28 | Epoch AI Benchmarking Hub |
| LiveBench | – | – | – | – | 74.2%#33 | LiveBench |
| Vals Indexnot in index | – | – | 53.6#23 | – | – | Vals AI |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Baidu Qianfan | 143 tok/s | 0.68 s | $0.440 | $1.32 | 1.0M | fp8 |
| Reka AI | 141 tok/s | 0.59 s | $0.110 | $0.660 | 262k | fp4 |
| Wafer | 124 tok/s | 0.45 s | $0.100 | $0.250 | 1.0M | – |
| NovitaAI | 81 tok/s | 1.60 s | $0.409 | $1.23 | 1.0M | fp8 |
| Fireworks | 79 tok/s | 0.82 s | $0.220 | $0.660 | 1.0M | – |
| Alibaba Cloud Int. | 65 tok/s | 1.04 s | $0.176 | $0.528 | 1M | – |
| Makora | 64 tok/s | 0.48 s | $0.090 | $0.195 | 1M | – |
| CoreWeave | 64 tok/s | 0.44 s | $0.130 | $0.280 | 262k | fp8 |
| Inceptron | 60 tok/s | 0.53 s | $0.130 | $0.350 | 1.0M | fp4 |
| Together | 53 tok/s | 0.68 s | $0.140 | $0.280 | 1.0M | – |
| GMICloud | 47 tok/s | 2.24 s | $0.286 | $0.858 | 1.0M | fp4 |
| AtlasCloud | 43 tok/s | 1.90 s | $0.440 | $1.32 | 1.0M | fp4 |
| SiliconFlow | 42 tok/s | 1.65 s | $0.220 | $0.660 | 1.0M | fp8 |
| StreamLake | 40 tok/s | 2.04 s | $0.088 | $0.264 | 1.0M | fp8 |
| Parasail | 39 tok/s | 0.61 s | $0.140 | $0.280 | 1.0M | fp8 |
| Cloudflare | 37 tok/s | 1.48 s | $0.440 | $1.32 | 1.3M | – |
| Baseten (US) | 36 tok/s | 1.77 s | $0.130 | $0.260 | 1.0M | fp8 |
| NextBit | 36 tok/s | 2.13 s | $0.352 | $1.06 | 1.0M | fp8 |
| Sail Research | 30 tok/s | 1.46 s | $0.065 | $0.180 | 1.0M | fp4 |
| Baseten | 29 tok/s | 1.55 s | $0.130 | $0.260 | 1.0M | fp8 |
| Phala | 28 tok/s | 0.92 s | $0.440 | $1.32 | 1.0M | – |
| Mancer | 26 tok/s | 0.74 s | $0.165 | $0.500 | 1.0M | fp8 |
| Venice | 25 tok/s | 0.98 s | $0.175 | $0.350 | 1M | – |
| Relace | 23 tok/s | 1.41 s | $0.065 | $0.180 | 1.0M | fp4 |
| DigitalOcean | 23 tok/s | 0.55 s | $0.080 | $0.252 | 1.0M | – |
| OpenInference | 21 tok/s | 1.64 s | $0.050 | $0.160 | 1.0M | fp8 |
| DeepInfra | 21 tok/s | 0.84 s | $0.060 | $0.180 | 1.0M | fp8 |
| Morph | 4 tok/s | 1.94 s | $0.123 | $0.347 | 1.0M | bf16 |
Price history
Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.040 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0001 | $0.0001 | 9.0 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0009 | $0.0007 | 17.1 s |
| Code edit | 6,000 / 1,500 | $0.0007 | $0.0005 | 41.4 s |
| Agentic coding session | 60,000 / 4,000 | $0.0046 | $0.0035 | 1.8 min |
| Structured extraction | 2,000 / 200 | $0.0002 | $0.0001 | 6.2 s |
See also
Data as of 10 Sept 2026. Compare these configurations.