AI Model Release Tracker

DeepSeek-V4.1-FlashOpen Weight

Released

DeepSeek-V4.1-Flash is an AI model released by DeepSeek on Thursday, Sep 10 2026, 28 days after DeepSeek-V4-Pro-0813. It is an open-weight model — the trained weights are available to download and run. It has 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 Terminal-Bench 3.0, Terminal-Bench 2.1, CyberGym, BullshitBench v2, DeepSWE 1.1, NL2Repo-Bench, and 7 more.

Available from

Decart——$0.09$0.181Mfp4
Baidu——$0.0537$0.21481Mfp8
Sail Research——$0.09$0.251Mfp4
Sail Researchus—$0.09$0.251Mfp4
DeepInfra—flex$0.112$0.3361Mfp8
DeepInfra——$0.14$0.421Mfp8
Relace——$0.07$0.501M—
AtlasCloud——$0.135$0.541Mfp8
StreamLake——$0.147$0.5881Mfp8
Morph——$0.454$0.601Mfp8
CoreWeave——$0.15$0.651Mfp8
Wafer——$0.19$0.651Mmxfp4
DigitalOcean——$0.165$0.661M—
ModelRun——$0.17$0.671Mfp4
InferenceNet——$0.103$0.7151Mfp8
InferenceNetus—$0.103$0.7151Mfp8
GMICloud——$0.18$0.721Mfp8
Reka——$0.105$0.841M—
Fireworks—spot$0.21$0.841M—
OpenInference——$0.02$1.001Mfp4
Mancer 2——$0.20$1.001Mfp8
Makora——$0.26$1.171Mfp8
Fireworks——$0.30$1.201M—
Crusoe——$0.29$1.201Mfp8
BaseTen——$0.30$1.201M—
BaseTenus—$0.30$1.201M—
Cohere——$0.30$1.201M—
Modal——$0.30$1.201M—
Nebius——$0.30$1.201Mfp8
Parasail——$0.30$1.201Mfp8
PrimeIntellect——$0.30$1.201M—
SiliconFlow——$0.30$1.201Mfp8
Together——$0.30$1.201M—
Venice——$0.30$1.201Mfp8
Alibaba——$0.33$1.321M—
Fireworksus—$0.45$1.801M—
BaseTen—fast$0.60$2.401M—
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 24, 2026Third-party rates from openrouter.ai, not the lab’s own list price.

Benchmarks

Coding

DeepSWE 1.1Agentic coding — Artificial Analysis' independent test of deep, agentic software-engineering work — the AI has to plan and carry out substantial coding tasks end to end. (Version 1.1 of the test.) Higher is better.
74.2%
#8 of 37
NL2Repo-BenchRepo-level code generation — Tests whether the AI can turn a natural-language requirement into working code across an entire repository, not just produce a single function or patch. Higher is better.
65.4%
#2 of 7

Terminal & CLI

Terminal-Bench 3.0Agentic terminal coding — command-line task completion (v3.0, much harder task set)
30%
#1
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.
90.6%
#1
Terminal-Bench 4.0Agentic terminal coding — Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Version 4.0 recalibrated how much time, CPU and memory each task gets, removed eight tasks and fixed nineteen, so fewer runs fail for reasons that have nothing to do with the model. Scores are not comparable with earlier versions. Higher is better.
31.2%
#18 of 30

Agentic & tool use

CyberGymCybersecurity — Tests the AI on cybersecurity challenges — finding and exploiting software weaknesses inside a safe sandbox. Higher is better.
88.1%
#1
Agent's Last ExamAgentic computer use — A hard set of desktop and operating-system tasks an AI agent has to finish by looking at the screen and working the machine itself. The score is the share it passes outright — partial credit does not count. Higher is better.
31.8%pass@1
#5 of 10
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.
54.8%
#3 of 19

Reasoning & science

Humanity's Last ExamMultidisciplinary reasoning — Humanity's Last Exam — extremely hard expert questions across many subjects, written so you can't just look up the answer. “No tools” means the AI answers on its own. Higher is better.
36.8%no tools
63.9%with tools
no tools
#15 of 25
with tools
#7 of 54
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better.
90.9%
#15 of 67

Multimodal

ChartographyChart tasks — A chart-centred test run with tools available to the AI, reported separately from the chart-reading benchmarks above it. Higher is better.
78.9%with tools
#6 of 7

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.
57%
#29 of 98

Community preference

threejsevalCommunity preference (Three.js) — Every model gets the same prompt — "the Eiffel Tower", "a glass fishbowl", "a robot arm picking toys into a box" — and builds a 3D scene in Three.js. Real people then see two scenes side by side, names hidden, and vote for the one they prefer. The votes become a chess-style Elo rating on threejseval.com, averaged across all the prompts. It measures whether the scene looks and moves right to a human eye, not whether the code passes a test. Higher is better.
1642
#9 of 29

Compare DeepSeek-V4.1-Flash with

DeepSeek-V4.1-Flash

Suggested comparisons

DeepSeek-V4.1-FlashvsDeepSeek-V4.1-Flash-1017DeepSeek-V4.1-FlashvsGPT-6.1 SolDeepSeek-V4.1-FlashvsClaude Haiku 5.5DeepSeek-V4.1-FlashvsGemini 4 ArgonDeepSeek-V4.1-FlashvsMuse Spark OpenDeepSeek-V4.1-FlashvsGrok 4.7DeepSeek-V4.1-FlashvsMistral Large 4DeepSeek-V4.1-FlashvsKimi K3.1 ProDeepSeek-V4.1-FlashvsGLM-5.3-FlashDeepSeek-V4.1-FlashvsQwen3.8-Max-0902DeepSeek-V4.1-FlashvsNemotron 3.5 Bolt E4B 4GB

About

DeepSeek V4.1 Flash was introduced in September 2026 with a new causal encoder-decoder architecture. DeepSeek reported 552 billion total parameters, with about eight billion active for input encoding and 16 billion active for output generation. The model was designed to improve inference speed and throughput while supporting native visual understanding.

The release brought this architecture to DeepSeek’s hosted assistant and API, with downloadable weights for self-managed deployment. Its vision capability belonged to the model rather than to a text-only endpoint supplemented solely by application tools. The 4.1 name also distinguishes it from the earlier V4 Flash checkpoints, including a separate experimental vision release in the V4 line.

Frequently asked questions

DeepSeek-V4.1-Flash was released by DeepSeek on Thursday, Sep 10 2026.

All DeepSeek releases

24 tracked