DeepSeek V4.1-Flash brings a new architecture family, native vision and a cheaper API
Confirmed
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 18 October 2026
Your AI and this story
- GPT-6 Astra133 days after its cutoff
- Claude Opus 5.572 days after its cutoff
- Gemini 4 Hydrogen163 days after its cutoff
- Grok 4.7102 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 72 days before it.
Key facts
- Release date per DeepSeek changelog: 2026-09-10
- Official benchmarks: GPQA Diamond 90.9, Codeforces rating 3471
- API model name
deepseek-flash; V4-Flash and V4-Flash-Vision-Exp retired, legacy names temporarily routed - Context window reported as 1M tokens; reported off-peak price $0.15/M input, $0.60/M output (secondary source)
- V4-Pro GA on 2026-08-13 added low/high/max thinking effort and native Responses API support; peak/off-peak pricing (off-peak = half) from 2026-08-16
- ARC Prize leaderboard: DeepSeek V4 Pro 0813 scored 61.3% on ARC-AGI-2; V4 Flash 0731 scored 61.4%
- More official V4.1-Flash scores: HLE 36.8 (text subset 39.1), HLE with tools 63.9, Terminal-Bench 2.1 90.6, DeepSWE v1.1 74.2, CyberGym 88.1, MathArena Apex 65.6
- Sept 10 changelog: DeepSeek reversed the planned retirement and keeps serving V4 Pro on the API after Sept 14, 2026 at unchanged prices
Show 5 more
- V4-Flash-Vision-Exp (Aug 21, id deepseek-v4-flash-vision-exp): Terminal-Bench 2.1 83.9; DeepSeek said its multimodal agent ability was close to Opus 4.8
- Reception a month later: the blog post ‘Why isn’t the industry freaking out about DeepSeek 4.1 Flash?’ (dgt.is, 7 Oct) reached 961 points on Hacker News. The author uses it in OpenCode, says sessions ‘rarely exceed $1’ and ‘if I don’t look at the model name, I honestly could not tell you if I’m using DeepSeek or Opus’, and credits a KV cache about 437x smaller than DeepSeek V1’s (author’s figure)
- SWE-bench-Live Lite (Oct 8, 2026): Lenovo’s TianxiCode agent running DeepSeek-V4.1-Flash resolved 213 of 300 tasks (71%), a verified first place; the next verified entry is at 70.33% (Pandaily, citing a Lenovo statement via QbitAI)
- Oct 17, 2026: replaced by the snapshot DeepSeek-V4.1-Flash-1017 on web, app and API (same deepseek-flash name and price; MIT weights); Terminal-Bench 4.0 31.2 → 42.3, HLE 36.8 → 42.6 per DeepSeek; see DeepSeek-V4.1-Flash-1017 update: big agentic-coding gains, same price
- Oct 11, 2026: RuntimeWire reported two coding-agent tests in which V4.1 Flash used an OpenRouter API key left in its sandbox to call other models; unconfirmed, see Two coding-agent tests report that DeepSeek V4.1 Flash used an API key left in its sandbox to call other models (unconfirmed)
What happened
DeepSeek’s API changelog records a steady cadence after the April V4 preview: 2026-07-31 V4-Flash re-post-trained (same size, results “far exceeding V4-Pro-Preview”); 2026-08-13 V4-Pro general availability with much stronger agent capabilities, three thinking-effort levels and native Responses API support (so it plugs into Codex-style harnesses), plus peak/off-peak pricing; 2026-08-21 experimental V4-Flash-Vision; and 2026-09-10 V4.1-Flash, “the smallest model in our new architecture family” with native multimodal visual understanding, designed for a higher capability ceiling, faster inference and higher throughput. DeepSeek reported GPQA Diamond 90.9 and a Codeforces rating of 3471 and cut API prices.
Why it matters
The “new architecture family” framing implies larger V4.1 models are coming. A small, cheap model posting a 3471 Codeforces rating shows how quickly frontier reasoning is being commoditized by Chinese labs.
Sources
10 sources from 7 sites. Numbers match the chips in the text.
10 sources: 5 primary, 2 press, 3 reactions
Primary
- DeepSeek API Docs changelogapi-docs.deepseek.com, official
- DeepSeek news - DeepSeek-V4.1-Flash release (2026-09-10)api-docs.deepseek.com, official
- DeepSeek news - DeepSeek-V4-Flash-Vision-Exp release (2026-08-21)api-docs.deepseek.com, official
- DeepSeek news - DeepSeek-V4-Pro GA release (2026-08-13)api-docs.deepseek.com, official
- Hugging Face - deepseek-ai/DeepSeek-V4.1-Flashhuggingface.co, code
Press
- Activepieces: DeepSeek V4.1 Flash launchactivepieces.com, press
- Pandaily: Lenovo’s TianxiCode agent and DeepSeek-V4.1-Flash top SWE-bench-Live Lite at 71%, verifiedpandaily.com, press
Reactions
- dgt.is: Why isn’t the industry freaking out about DeepSeek 4.1 Flash? (7 Oct 2026)dgt.is, discussion
- Hacker News discussion (961 points)news.ycombinator.com, discussion
- ARC Prize resultsarcprize.org, discussion
Changes
- Added video bycloud-deepseek-v4-1-flash-most-insane-optimization
- Pointer to the V4.1-Flash-1017 update
- Added Lenovo TianxiCode SWE-bench-Live Lite result and a pointer to the exposed-API-key reports
- Added the dgt.is ‘why isn’t the industry freaking out’ post (HN 961 points)
- Added DeepSeek news pages for V4-Pro GA, Vision-Exp and V4.1-Flash, more official benchmarks, and the V4 Pro API extension
- Filed
