Get started

DeepSeek V4 Flash 0731 (batch)

Sold by 26 sellers, timed at 115 tok/s, scored 69.1 on the Coding Index, on 8 Design Arena boards, and with results on 2 openly licensed boards.

retrieved sha256 · 0bb53a77713f

Models/DeepSeek/DeepSeek V4 Flash 0731 (batch)USD per 1M tokens · as OpenRouter routes it
deepseek/deepseek-v4-flash-0731:batchDeepSeek · batch variant · DeepSeek tokenizer
OpenRouter list price · the routed rate, as the catalog carried itretrieved 17 Nov 16:29 UTC from openrouter.ai ↗
Input$0.048per 1M tokens
Output$0.144per 1M tokens
Cached input$0.012per 1M tokens
Context1.05M1,048,576 tokens
What the list says · the fields OpenRouter publishes for this id12 fields
AuthorDeepSeek
Listed
Canonical iddeepseek/deepseek-v4-flash-20260731
Context1,048,576 tokens
Max output384,000 tokens
Modalitiestext → text
TokenizerDeepSeek
Reasoningoptional, on by default · effort max, high, low, none, high by default
Moderationnot moderated by OpenRouter
Variantbatch
Other listingbase listing
What the weights say · read from the model's own repository on Hugging Faceretrieved from huggingface.co ↗
Licencemit
Parameters304B · 304,180,418,494 declared
Model classDeepseekV4ForCausalLM
Downloadnot gated
Published

What sellers charge

26 sellers · 10x from cheapest to dearest

One tick per seller, on a log scale of input price per million from $0.044 at Baidu to $0.44 at PrimeIntellect. 2 sellers undercut the $0.048 listed; 23 are dearer. A faint tick declares no number format. Every seller is named on the endpoint ledger below.

The same sellers grouped by the number format each declares, on the same scale. The fp8 group alone spans 100% of the range, so the cheap end is not cheap for its number format. A declaration is checked by nobody: this groups the rows and certifies none.

Measured speed

810,400 requests · 26 endpoints · 30-minute window
Throughput115 tok/sMedian output rate, weighted across the 26 endpoints serving it by the requests each answered.the speed board
First token456msMedian wait before the first token, over the same window and weighted the same way.how Vitna reads a figure
Endpoint spread40 to 162The slowest and the fastest route serving this model, in tokens per second. Speed is a property of the route, not of the weights.the board, fastest first

OpenRouter's own timing of its routed traffic, weighted by requests, over one 30-minute window read . Every route is on the endpoint ledger below.

Every endpoint

26 priced · 26 timed · cheapest first
  1. Baidubaidu/fp8 · 90% off$0.044$0.132$0.0014fp8131k102 tok/s66742ms87,749 req100%
  2. Waferwafer/us$0.0443$0.30$0.035nvfp4944knot matchednonenonenot matched99.93%
  3. DeepInfradeepinfra/fp8/flex · listed$0.048$0.144$0.012fp8384knot matchednonenonenot matched72.50%
  4. DeepInfradeepinfra/fp8$0.06$0.18$0.015fp8384knot matchednonenonenot matched98.77%
  5. Relacerelace/fp4$0.07$0.14$0.014fp4944k112 tok/s51623ms23,943 req99.23%
  6. StreamLakestreamlake/fp8 · 80% off$0.088$0.264$0.0028fp8384k111 tok/s511.24s14,530 req100%
  7. DeepInfradeepinfra/fp8/priority$0.09$0.27$0.0225fp8384knot matchednonenonenot matched99.30%
  8. Coherecohere/fp8$0.11$0.22$0.05fp8384k162 tok/s56484ms45,956 req100%
  9. DigitalOceandigitalocean$0.119$0.238$0.0238not declared944k152 tok/s65680ms23,575 req100%
  10. BaseTenbaseten/fp32$0.13$0.26$0.028fp32384knot matchednonenonenot matched100%
  11. BaseTenbaseten/fp8$0.13$0.26$0.028fp8384knot matchednonenonenot matched100%
  12. CoreWeavecoreweave/fp8$0.13$0.27$0.05fp8236k119 tok/s81234ms181,863 req100%
  13. Inceptroninceptron/fp4$0.13$0.495$0.027fp4944k82 tok/s36699ms1,604 req98.47%
  14. Venicevenice$0.13$0.26$0.028not declared262k93 tok/s59586ms2,937 req99.71%
  15. Nebiusnebius/fp8$0.14$0.28$0.01fp8944k152 tok/s571.09s13,458 req100%
  16. Parasailparasail/fp8$0.14$0.28$0.05fp8944k103 tok/s54392ms15,157 req99.79%
  17. Waferwafer/fast$0.14$0.23$0.135nvfp4944knot matchednonenonenot matched99.86%
  18. Mancer 2mancer/fp8$0.15$0.50$0.05fp8944k74 tok/s45.3822ms164 req98.76%
  19. Sail Researchsail-research/fp4$0.15$0.30$0.042fp4944k40 tok/s331.05s1,162 req100%
  20. Sail Researchsail-research/us$0.15$0.30$0.042fp4944k40 tok/s251.35s1,063 req99.30%
  21. Alibabaalibaba$0.176$0.528$0.0176not declared393knot matchednonenonenot matched100%
  22. Alibabaalibaba/us-east-1$0.187$0.561$0.0187not declared393knot matchednonenonenot matched100%
  23. SiliconFlowsiliconflow/fp8$0.22$0.66$0.028fp8393k55 tok/s29916ms289 req98.70%
  24. Novitanovita/fp8 · 5% off$0.418$1.254$0.0266fp8393k40 tok/s30828ms7,361 req99.93%
  25. Chuteschutes/fp8$0.44$1.32$0.044fp8131k92.5 tok/s36.11.14s120 req99.30%
  26. PrimeIntellectprimeintellect/fp4$0.44$1.32$0.014fp466k101 tok/s46422ms8 req100%
  27. Alibabanot matchednot matchednot matchednot matchedno offer104 tok/s381.61s25 reqno offer
  28. Alibabanot matchednot matchednot matchednot matchedno offer104 tok/s44.82.14s728 reqno offer
  29. BaseTennot matchednot matchednot matchednot matchedno offer158 tok/s72369ms11,073 reqno offer
  30. BaseTennot matchednot matchednot matchednot matchedno offer99 tok/s48264ms2,892 reqno offer
  31. DeepInfranot matchednot matchednot matchednot matchedno offer128 tok/s60407ms194,782 reqno offer
  32. DeepInfranot matchednot matchednot matchednot matchedno offer94 tok/s322.11s25 reqno offer
  33. DeepInfranot matchednot matchednot matchednot matchedno offer70 tok/s40461ms95,241 reqno offer
  34. Wafernot matchednot matchednot matchednot matchedno offer59 tok/s48405ms625 reqno offer
  35. Wafernot matchednot matchednot matchednot matchedno offer103 tok/s73333ms84,070 reqno offer

Each track runs from a route's p90, the rate nine in ten of its requests met or beat, up to its median. Weighted by requests, the figure above reports where the traffic went rather than what one route does: 20 of the 26 timed routes are slower than it, and the middle route runs at 102 tok/s. The widest median-to-p90 gap is DeepInfra's, 2.94x.

Alibaba publishes 2 offers and had 2 routes timed; BaseTen publishes 2 offers and had 2 routes timed; DeepInfra publishes 3 offers and had 3 routes timed; Wafer publishes 2 offers and had 2 routes timed. The two documents share no route id, so those offers and routes are printed apart rather than paired.

Uptime is OpenRouter's own figure for each seller over the trailing day, read once a day, and the column prints the latest reading, taken . 24 of the 35 sellers with a reading fell under 99% on at least one of the 23 days held, 23 Oct 2026 to 21 Nov 2026: Chutes at 0.00% on 6 Nov 2026, DeepInfra (deepinfra/fp8/flex) at 47.89% on 19 Nov 2026, SiliconFlow at 73.80% on 12 Nov 2026, Nebius at 83.14% on 12 Nov 2026, and 20 more. Source: OpenRouter's public endpoints API, one reading a day for each seller: 1,211 sellers carried one on the latest of the 23 days held.

Source: OpenRouter's public endpoints API. 1257 sellers across 359 models when the mirror ran, 186 of them served by more than one. List prices, not what anybody paid; a discount is the seller's own. Read . Speed is OpenRouter's own timing over one 30-minute window, read .

What each seller accepts

22 parameters · 26 sellers
  1. include_reasoningevery seller
  2. max_tokensevery seller
  3. reasoningevery seller
  4. reasoning_effortevery seller
  5. temperatureevery seller
  6. tool_choiceevery seller
  7. toolsevery seller
  8. top_pevery seller
  9. response_format24 of 26not BaseTen (baseten/fp32) and BaseTen (baseten/fp8)
  10. stop24 of 26not DigitalOcean and SiliconFlow
  11. frequency_penalty23 of 26not Baidu, DigitalOcean, and StreamLake
  12. presence_penalty22 of 26
  13. structured_outputs22 of 26
  14. top_k20 of 26
  15. seed18 of 26
  16. repetition_penalty14 of 26
  17. logit_bias11 of 26
  18. logprobs11 of 26
  19. top_logprobs11 of 26
  20. min_p8 of 26
  21. parallel_tool_calls3 of 26only Cohere, Inceptron, and PrimeIntellect
  22. top_a2 of 26only Mancer 2 and PrimeIntellect

8 of these 22 are accepted by every one of the 26 sellers that publish a list; the other 14 are declared by some sellers and not others, so what a request sending one of them gets depends on which seller serves it. Each list is the seller's own declaration in OpenRouter's endpoints API, read , and nothing here tested one.

Price position

342 priced models

Blended $0.072 per million tokens, 3 in to 1 out: cheaper than 324 of the 342 priced models in the catalog. Each bar is one stretch of the catalog on a logarithmic price scale; the mark is this listing.

Measured quality

145 models with an index · 132 on a design arena board

Each mark is one board this model stands on, placed at the share of that board it scores above, and led to its number in the list under the rule. The columns and the ledger below carry the figures.

Intelligence Indexnot scoredArtificial Analysis has not scored this model on its Intelligence Index.the intelligence leaderboard
Coding Index69.1Position 27 of 118, above 77% of the models scored.the coding leaderboard
Agentic Index41.0Position 21 of 94, above 78% of the models scored.the agentic leaderboard
  1. 3D scenes · Models1188 Elowin rate 48.8% · source rank 56the board, 113 ranked
  2. ASCII art · Models1082 Elowin rate 34.0% · source rank 66the board, 67 ranked
  3. Code · Models1217 Elowin rate 46.5% · source rank 59the board, 120 ranked
  4. Data visualisation · Models1169 Elowin rate 40.0% · source rank 71the board, 110 ranked
  5. Game development · Models1197 Elowin rate 44.8% · source rank 59the board, 120 ranked
  6. SVG · Models1169 Elowin rate 44.7% · source rank 44the board, 83 ranked
  7. UI components · Models1225 Elowin rate 47.1% · source rank 52the board, 116 ranked
  8. Websites · Models1229 Elowin rate 46.8% · source rank 58the board, 126 ranked
Epoch Capabilities Index · a fit by a third party over every benchmark this model has results onthe whole index, with intervals
Index154.4152.6 to 156.4
Position53 of 258above 80% of the models placed
Results behind it15benchmark results
As Epoch names itDeepSeek V4 Flash 0731released 31 Jul 2026

Position 53 of the 258 models Epoch has placed, which is not the population of either board above, on a scale anchored at Claude 3.5 Sonnet 130 and GPT-5 150. Epoch states that absolute values are meaningless by themselves and that comparisons between models are what the index is for, that values are not linearly related to benchmark accuracy, and that a highly specialised model may score low while being capable in its own domain.

Benchmark results

2 of 6 boards · 30 results
Epoch AI Benchmarking Hub · publisher-run, CC-BY-4.07 of 22 tasks · at max effort

Mean accuracy, 0 to 1, on the task the row names, as Epoch AI runs it and averaged over repeated runs. All eighteen of their tasks, including the held-out FrontierMath tiers, named as they name them. The model id carries the effort setting the run used. Epoch AI Benchmarking Hub names this run deepseek-v4-flash-0731_max. All runs is how many runs the publisher scored on the task, every model's. Read from epoch.ai ↗, which dates its own data . Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.

LiveBench · publisher-run, Apache-2.023 of 23 tasks · 71 runs scored on each task · effort unstated

Percent correct on the task the row names, higher is better. All 23 published columns; there is no average to carry, because the site computes its headline in the browser from a weighting this does not reimplement. LiveBench names this run deepseek-v4-flash-0731, with no effort setting written on it. Read from livebench.ai ↗, which dates its own data . LiveBench, Apache-2.0. Results produced by the LiveBench maintainers on questions refreshed monthly.

2 of the 6 openly licensed boards Vitna mirrors carry a result for DeepSeek V4 Flash 0731 (batch), each printed as its publisher published it and none added to another. MTEB, HELM, Vectara hallucination leaderboard and Aider polyglot have not published one under a name that matches its listing, and names are matched to the catalog exactly and never guessed. A task that is a link opens its column on /capability, with every run scored on it; an effort that is a link opens its run's row there, where the run is ranked on Vitna's scale. Every board, with its licence and what its publisher measures, is on /benchmarks.

Sources

Source: OpenRouter (openrouter.ai/models), as of 17 Nov 16:29 UTC. sha256 0bb53a77713f over 853,270 bytes.
Catalog: OpenRouter's public model list.
Intelligence, coding and agentic indexes: Artificial Analysis, as published by OpenRouter.
Elo, win rate and rank: Design Arena, as published by OpenRouter.
Source: each model's own Hugging Face repository (huggingface.co), as of 20 Nov 2026. 162 repositories read, 4 did not answer.
Licence, parameter count, model class and gating: each model's own Hugging Face repository, the one OpenRouter names for the listing.
A licence is the tag the uploader put on the repository, read from nothing else and verified by nobody.
A parameter count is the total the repository's safetensors files declare; for a mixture of experts that is every expert, not the share one token runs through.
Source: the Epoch AI Capabilities Index (epoch.ai/eci), as of 20 Nov 2026. 258 models placed over 2917 benchmark results.
Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
Vitna ran none of these evaluations and adjusted none of these figures.