DeepSeek V4 Pro 0813

Aug 2026
DeepSeek V4 Pro 0813
Compare in
on
$0.36per task?
$1.32 / $3.96per 1M in / out?cache $0.044
137 tok/sfirst token in 1.5 s?
1Mcontext

Rankingin non-thinking mode?

Full ranking →
1Fable 5.11,560.12Opus 51,515.43Fable 51,506.44Gemini 3.8 Flash1,506.35GPT-5.51,498.5
···
20Muse Spark 1.11,447.821DeepSeek V4 Pro 08131,445.622Kimi K2.61,443.2

Strong at

S
Science#14 of 312
B
Coding#51 of 132Agents & work tasks#71 of 175Knowledge#102 of 318
C
Reasoning#181 of 334Long context#175 of 273

Alternatives?

ModelScoreCostPer taskWhat changes
CheaperGemini 3.8 Flashhighwins 59%-24%*$1.24Sampling parameters are ignored+1BetterGPT-5.6 Solxhighwins 54%+304%$1.18OpenKimi K3maxwins 56%+203%$2.00Memory (4-bit): 990 GB → 1,668 GB+4CheaperGemini 3 Flash Previewunstatedwins 56%-43%$0.50 / $3CheaperGemini 3.6 Flashhighwins 54%-24%*$1.60

Run it yourself

estimate
Fits on✗ 8 GB (phone, laptop)✗ 16 GB✗ 24 GB GPU✗ 48 GB GPU✗ 80 GB GPU
LicenceMITfree for commercial use

Benchmarks

in Chat & Knowledge
AA Intelligence Index 4.3.220.4#119
AA Omniscience Index-12.4#102
AA Omniscience · Accuracy41.5#43
AA Omniscience · Hallucination Rate91.9#274
Humanity's Last Exam10.6#182
AA-LCR (long context) 1.150.7#177
CritPt (physics)0#179
GDP.pdf · AA all-pass3.6#96
GPQA Diamondlab92.4
HLE text-onlylab42.7
MathArena Apexlab65.3
SEC-Bench Prolab56.4
ExploitGymlab5.4
MMLU-Pro EMlab87.5
SimpleQA-Verifiedlab57.9
Chinese-SimpleQAlab84.4
HMMT February 2026lab95.2
IMOAnswerBenchlab89.8
Apexlab38.3
Apex Shortlistlab90.2
MRCR 1M MMRlab83.5
CorpusQA 1Mlab62
GDPval-AA Elolab1,554

Watch out 5

Pass back reasoning_content · Chat Completions›

With thinking and tools, return all prior reasoning_content even on turns without tool calls or the API returns HTTP 400. Source ↗

Disable thinking mode first · Chat Completions›

Required and named tool choices return HTTP 400 in thinking mode; disable thinking first. Source ↗

Parameters have no effect›

In thinking mode, temperature, presence_penalty and frequency_penalty have no effect. Source ↗

top_p value is ignored›

In thinking mode, top_p is fixed at 0.95 with tools and 1.0 otherwise; in non-thinking mode it takes effect. Source ↗

Stateless API · Responses›

Responses does not support previous_response_id or conversation; preserve conversation history in your client. Source ↗

FAQ

How much does DeepSeek V4 Pro 0813 cost?›

$1.32 per 1M input tokens and $3.96 per 1M output tokens; cached input $0.044 per 1M; cheapest via Sail Research at $0.90 / $0.58. A typical benchmark task costs about $0.36 at default effort.

Is DeepSeek V4 Pro 0813 free to use?›

There is no free API; the cheapest paid price is $0.90 per 1M input tokens. The weights are open, so you can run it yourself.

How good is DeepSeek V4 Pro 0813, and what is it best at?›

#181 of 334 in Chat & Knowledge, #51 of 132 in Coding, #48 of 133 in Agents & Tool Use, #30 of 195 in Writing & Language, #96 of 133 in Vision & Documents. Strongest at: Science.

Is DeepSeek V4 Pro 0813 better than Muse Spark 1.1?›

On Arena Text · Japanese, DeepSeek V4 Pro 0813 scores 1,445.6 vs 1,447.8 — about the same.

How can I use DeepSeek V4 Pro 0813?›

Through the API of Alibaba, AtlasCloud, Azure, Azure AI Foundry, Cloudflare and 13 more.

What is the context window of DeepSeek V4 Pro 0813?›

1M tokens, with up to 384K output tokens.

When was DeepSeek V4 Pro 0813 released? Is there a newer version?›

Released 2026-08-13. It is the current version in its line.

Is DeepSeek V4 Pro 0813 open source?›

Yes, the weights are open (MIT licence).

Can I run DeepSeek V4 Pro 0813 locally?›

The weights are open and can be self-hosted; hardware needs are not confirmed yet.

Can DeepSeek V4 Pro 0813 read images?›

No, it accepts text only.

What is DeepSeek V4 Pro 0813?›

DeepSeek V4 Pro 0813 is an open-weights language model from DeepSeek, released Aug 2026, the successor to DeepSeek V4 Pro 0424.

What are the best alternatives to DeepSeek V4 Pro 0813?›

Gemini 3.8 Flash (cheaper), GPT-5.6 Sol (better), Kimi K3 (open weights).