DeepSeekDeepSeek V4 Flash
DeepSeek V4 Flash is a DeepSeek open-weights reasoning model, released 24 Apr 2026. Its best configuration (max reasoning effort) ranks #100 of 751 on the BenchLeader Index at 60.7 ±4.9, in the upper half. The ± is the point: 192 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (77) and lowest in coding (48). At $0.262 per million tokens blended it is cheaper than most ranked models. Output speed of 216 tokens per second puts it in the fastest quarter, with a first answer in 10.2 s. It has been superseded by DeepSeek V4 Flash 0731, which is the version to compare against today. It has been measured at 4 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 24 Apr 2026.
- Blended price
- $0.262/M
- $0.150 in · $0.600 out
- Output speed
- 216 tok/s
- measured by Artificial Analysis
- First answer
- 10 s
- first token 1.35 s
- Context
- 1M
- Full answer
- 13 s
- median, reasoning included
- Cost per run
- $0.220
- one full Intelligence Index run
- Released
- 24 Apr 2026
- Overall index61
- Reasoning69
- Coding48
- Agents & tools67
- Maths57
- Knowledge55
- Instruction following77
- Long context65
- Composite69
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 0731high | 31 Jul 2026 | 54.4 | #235 |
| DeepSeek V4 Flashmaxthis page | 24 Apr 2026 | 60.7 | #100 |
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. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | 49.2 | #385 | 54 tok/s | 1.35 s | $0.0002 | Agents & tools 63 · Composite 51 · Instruction following 49 · Knowledge 42 · Long context 46 · Reasoning 42 |
| high | 57.6 | #165 | 54 tok/s | 1.35 s | $0.0002 | Agents & tools 59 · Coding 55 · Composite 57 · Human preference 63 · Instruction following 72 · Knowledge 51 · Long context 61 · Maths 63 · Reasoning 52 |
| maxbest | 60.7 | #100 | 216 tok/s | 10 s | $0.0002 | Agents & tools 67 · Coding 48 · Composite 69 · Instruction following 77 · Knowledge 55 · Long context 65 · Maths 57 · Reasoning 69 |
| not stated | 51.2 | #332 | 54 tok/s | 1.35 s | $0.0002 | Coding 62 · Composite 22 · Human preference 62 · Maths 61 · Reasoning 52 |
How its index has moved
8 Sept 2026 to 3 Oct 2026- max
- high
- not stated
- no reasoning
The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
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 | high | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| SimpleBench | – | – | – | 46.3%#52 | SimpleBench | |
| LMArena Hard Prompts | – | 1458#99 | – | 1459#95 | LMArena | |
| LiveBench Reasoningnot in index | – | – | – | 70.6%#61 | LiveBench | |
| GPQA Diamond (AA)not in index | 71.6%#279 | 86.7%#104 | 90.8%#55 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 7.8%#334 | 30.3%#131 | 38.5%#89 | – | Artificial Analysis | |
| Kagi LLM Benchmark | – | – | – | 52.2%#76 | Kagi LLM Benchmark | |
| CritPt | 0.3%#250 | 3.4%#156 | 16.6%#76 | – | Artificial Analysis | |
| LMCAnot in index | – | – | 35.9%#103 | – | Epoch AI Benchmarking Hub | |
| DTBenchnot in index | – | – | 86.4%#87 | – | Epoch AI Benchmarking Hub |
Coding
| Benchmark | no reasoning | high | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| SciCode | – | 42.0%#130 | 44.9%#117 | – | SciCode | |
| WeirdML | – | 43.8%#86 | 45.6%#83 | – | WeirdML | |
| LMArena Coding | – | 1480#103 | – | 1484#98 | LMArena | |
| LMArena WebDev | – | 1581#30 | – | – | LMArena | |
| LiveBench Codingnot in index | – | – | – | 69.2%#59 | LiveBench | |
| SciCode (AA)not in index | – | 40.2%#154 | 50.4%#96 | – | Artificial Analysis | |
| ALE-Benchnot in index | 325.0#107 | 678.2#78 | – | – | Epoch AI Benchmarking Hub | |
| Terminal-Bench 4.0 (AA)not in index | – | 3.0%#94 | 12.1%#71 | – | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | – | 56.9%#99 | 78.7%#49 | – | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | high | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| LiveBench Agentic Codingnot in index | – | – | – | 37.6%#62 | LiveBench | |
| Terminal-Bench Hard | 34.1%#90 | 38.6%#61 | 35.6%#74 | – | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 94.4%#31 | 95.6%#20 | 95.0%#27 | – | Artificial Analysis | |
| GDPval (AA)not in index | – | 24.7%#143 | 46.3%#59 | – | Artificial Analysis | |
| τ²-Bench Banking (AA) | – | 26.2%#75 | 39.4%#33 | – | Artificial Analysis | |
| ITBench SRE (AA)not in index | – | – | 31.5%#31 | – | Artificial Analysis | |
| Analyst Agent (AA)not in index | – | – | 25.0%#22 | – | Artificial Analysis |
Maths
| Benchmark | no reasoning | high | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| LiveBench Mathematicsnot in index | – | – | – | 79.7%#57 | LiveBench | |
| AIME 2026 | – | – | 95.8%#13 | – | MathArena | |
| HMMT February 2026 | – | – | 93.9%#10 | – | MathArena | |
| MathArena Apex | – | – | 27.1%#11 | – | MathArena | |
| LMArena Maths | – | 1441#86 | – | 1425#108 | LMArena |
Knowledge
| Benchmark | no reasoning | high | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| LiveBench Data Analysisnot in index | – | – | – | 68.0%#51 | LiveBench | |
| AA-Omniscience | -44.0#323 | -23.1#238 | -14.3#200 | – | Artificial Analysis | |
| AA-Omniscience: accuracynot in index | 26.0%#203 | 34.9%#144 | 40.4%#116 | – | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 5.5%#495 | 11.0%#378 | 8.3%#442 | – | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | high | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| LiveBench Languagenot in index | – | – | – | 70.1%#60 | LiveBench | |
| LiveBench Instruction Followingnot in index | – | – | – | 63.1%#49 | LiveBench | |
| IFBench | 47.2%#188 | 73.5%#40 | 79.2%#11 | – | Artificial Analysis | |
| LMArena Instruction Followingnot in index | – | 1434#89 | – | 1428#93 | LMArena |
Human preference
| Benchmark | no reasoning | high | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| LMArena Text | – | 1438#95 | – | 1436#100 | LMArena | |
| LMArena Creative Writingnot in index | – | 1410#88 | – | 1408#92 | LMArena | |
| LMArena Multi-turnnot in index | – | 1446#95 | – | 1453#83 | LMArena | |
| LMArena Longer Queriesnot in index | – | 1449#90 | – | 1448#91 | LMArena |
Long context
| Benchmark | no reasoning | high | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| AA-LCR | 41.7%#329 | 72.0%#177 | 79.7%#80 | – | Artificial Analysis |
Composite
| Benchmark | no reasoning | high | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | – | 146.1#70 | Epoch AI Benchmarking Hub | |
| LiveBench | – | – | – | 65.5%#60 | LiveBench | |
| AA Intelligence Index | 18.9#228 | 24.4#159 | 34.3#82 | – | Artificial Analysis |
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 |
|---|---|---|---|---|---|---|
| Alibaba Cloud Int. | 93 tok/s | 0.83 s | $0.134 | $0.268 | 1M | fp8 |
| Relace | 84 tok/s | 0.68 s | $0.022 | $1.28 | 1.0M | fp4 |
| Azure | 84 tok/s | 0.99 s | $0.210 | $0.560 | 1.0M | – |
| Baidu Qianfan | 73 tok/s | 0.84 s | $0.140 | $0.280 | 1.0M | fp8 |
| Cloudflare | 71 tok/s | 1.26 s | $0.440 | $1.32 | 384k | – |
| AtlasCloud | 70 tok/s | 1.62 s | $0.140 | $0.280 | 1.0M | fp4 |
| StreamLake | 66 tok/s | 1.45 s | $0.028 | $0.056 | 1.0M | fp8 |
| Parasail | 61 tok/s | 0.81 s | $0.140 | $0.280 | 1.0M | fp8 |
| NovitaAI | 60 tok/s | 1.37 s | $0.140 | $0.280 | 1.0M | fp8 |
| GMICloud | 55 tok/s | 1.55 s | $0.091 | $0.182 | 1.0M | fp8 |
| DeepInfra | 51 tok/s | 0.82 s | $0.090 | $0.180 | 1.0M | fp8 |
| Venice | 49 tok/s | 1.02 s | $0.097 | $0.193 | 1M | – |
| SiliconFlow | 42 tok/s | 1.16 s | $0.130 | $0.280 | 1.0M | fp8 |
| Open Inference | 37 tok/s | 0.85 s | $0.029 | $1.66 | 1.0M | fp4 |
| Mancer | 29 tok/s | 0.82 s | $0.190 | $0.500 | 1.0M | fp8 |
| DigitalOcean | 6 tok/s | 1.78 s | $0.098 | $0.196 | 1.0M | – |
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 $0.003 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0002 | $0.0002 | 11.6 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0022 | $0.0008 | 13.0 s |
| Code edit | 6,000 / 1,500 | $0.0018 | $0.0011 | 17.2 s |
| Agentic coding session | 60,000 / 4,000 | $0.011 | $0.0048 | 28.7 s |
| Structured extraction | 2,000 / 200 | $0.0004 | $0.0002 | 11.1 s |
See also
Data as of 4 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “DeepSeek V4 Flash: benchmarks, pricing, speed and rank”, https://www.benchleader.com/models/deepseek-v4-flash, data as of 4 Oct 2026.