AI Model Release Tracker

DeepSeek-V4-Flash-0731Open Weight

Released

DeepSeek-V4-Flash-0731 is an AI model released by DeepSeek on Friday, Jul 31 2026, 98 days after DeepSeek-V4-Flash. It is an open-weight model — the trained weights are available to download and run. It is a 284B parameterThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. model with a 1M token context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider.. Benchmark results (shown below) cover BullshitBench v2, SWE-Bench Verified, Terminal-Bench 2.1, Toolathlon-Verified, BrowseComp, CyberGym, and 3 more.

Available from

Relace——$0.057$0.13411Mfp4
DeepInfra——$0.06$0.181Mfp8
Wafer—fast$0.10$0.231Mnvfp4
DigitalOcean——$0.119$0.2381M—
BaseTen——$0.13$0.261Mfp32
BaseTen——$0.13$0.261Mfp8
Venice——$0.13$0.261M—
Reka——$0.10$0.28256K—
CoreWeave——$0.13$0.28256Kfp8
Cohere——$0.14$0.281M—
Parasail——$0.14$0.281Mfp8
Sail Research——$0.15$0.301Mfp4
Sail Researchus—$0.15$0.301Mfp4
Inceptron——$0.13$0.36881Mfp4
StreamLake——$0.132$0.3961Mfp8
Mancer 2——$0.20$0.601Mfp8
SiliconFlow——$0.22$0.661Mfp8
GMICloud——$0.286$0.8581Mfp8
Phala——$0.308$0.9241M—
Alibaba——$0.352$1.0561M—
Alibabaus-east-1—$0.374$1.1221M—
Novita——$0.4092$1.22761Mfp8
AtlasCloud——$0.44$1.321Mfp4
Baidu——$0.44$1.321Mfp8
Cloudflare——$0.44$1.321M—
USD per 1M tokensEvery price here is for one million tokens. A token is roughly three-quarters of a word, so a million tokens is about 750,000 words of text.Checked October 29, 2026Third-party rates from openrouter.ai, not the lab’s own list price.

Benchmarks

Coding

SWE-Bench VerifiedCoding — Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better.via BenchLM
79%
#16 of 70

Terminal & CLI

Terminal-Bench 2.1Agentic terminal coding — Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Higher is better.
82.7%
#17 of 36

Agentic & tool use

Toolathlon-VerifiedPersonal tool use — Tests how well the AI uses everyday personal tools and apps to get things done — a human-checked version of Toolathlon. Higher is better.
70.3%
#8 of 9
BrowseCompWeb browsing — Can the AI browse the web and track down hard-to-find answers? Higher is better.via BenchLM
73.2%
#25 of 33
CyberGymCybersecurity — Tests the AI on cybersecurity challenges — finding and exploiting software weaknesses inside a safe sandbox. Higher is better.
76.7%
#12 of 13
AutomationBenchBusiness workflows — Tests whether the AI can run real multi-step business workflows — the kind of end-to-end office processes companies want to automate — from start to finish. Higher is better.
25.1%
#20 of 20

Reasoning & science

Humanity's Last ExamMultidisciplinary reasoning — Humanity's Last Exam — extremely hard expert questions across many subjects. “With tools” means the AI is allowed to search the web or run code while answering. Higher is better.via BenchLM
34.8%with tools
#38 of 54
ARC-AGI-2Abstract reasoning — Puzzle-style tests of abstract reasoning and pattern-finding — the kind of thing people find easy but AIs often struggle with. Higher is better.via BenchLM
61.4%
#23 of 41

Robustness

BullshitBench v2Nonsense detection — Given a confidently-worded but nonsensical prompt, does the AI spot that it makes no sense and push back — instead of playing along and inventing an answer? The score is how often it clearly called out the nonsense. Higher is better.
39%
#50 of 98

Source: BenchLM, retrieved 28 September 2026. Other sources are identified on the linked benchmark pages.

Compare DeepSeek-V4-Flash-0731 with

DeepSeek-V4-Flash-0731

Suggested comparisons

DeepSeek-V4-Flash-0731vsDeepSeek-V4.1-Flash-1017DeepSeek-V4-Flash-0731vsGPT-6.1 TerraDeepSeek-V4-Flash-0731vsClaude Haiku 5.5DeepSeek-V4-Flash-0731vsGemini 4 CarbonDeepSeek-V4-Flash-0731vsMuse Spark OpenDeepSeek-V4-Flash-0731vsGrok 4.7DeepSeek-V4-Flash-0731vsMistral Large 4DeepSeek-V4-Flash-0731vsKimi K3.1 ProDeepSeek-V4-Flash-0731vsGLM-5.3-FlashDeepSeek-V4-Flash-0731vsQwen3.8-Max-0902DeepSeek-V4-Flash-0731vsNemotron 3.5 Bolt E4B 4GB

About

DeepSeek V4 Flash 0731 was the updated V4 Flash checkpoint that DeepSeek’s model card described as superseding the preview version. DeepSeek announced it on July 31, 2026 as the official V4-Flash API, in public beta under the unchanged deepseek-v4-flash name, with native Responses API support adapted for Codex. The card emphasized agentic tasks including software engineering and security work. It documented a model structure shared with the V4-Flash-DSpark build, including its speculative-decoding module.

DeepSeek published this checkpoint’s weights under the MIT License. The model card reported 82.7% on Terminal-Bench 2.1 using a minimal DeepSeek Harness at maximum reasoning effort; that is a result for the stated model-and-agent setup. The dated checkpoint name distinguishes it from both the April V4 Flash preview and the later V4.1 Flash architecture.

Frequently asked questions

DeepSeek-V4-Flash-0731 was released by DeepSeek on Friday, Jul 31 2026.

All DeepSeek releases

24 tracked