DeepSeek V4.1 Flash
DeepSeek's fast, low-cost open-weight model (MIT license) with a 1M-token context window, and the model behind DeepSeek's deepseek-flash API. DeepSeek published an updated V4.1-Flash-1017 snapshot on October 17, 2026, after the benchmark runs recorded here.
DeepSeekLanguage modelReleased September 2026Data checked October 18, 2026
This model is not in the current people's rankings: it is either new and still collecting votes, or ranked below the top 12. See the AI leaderboard for the models people rate highest.
On paper
Fixed-test rankings from the benchmarks tab: Speed #3, Reasoning #19.
Strengths and weaknesses
- Very fast (176 tokens per second on Artificial Analysis) and cheap ($0.30 / $1.20 per million tokens at peak, half that off-peak)
- 72.9% on ARC-AGI-2, the best score among the open models here
- Just under the 40 bar on the Artificial Analysis Intelligence Index (39.5) at $0.27 per task
- Weaker on hard reasoning than the frontier models (39.2% on Humanity's Last Exam)
- Benchmark results here predate the October 17 update, so the current API model may score differently
Best for
- A very cheap all-round assistant
- Fast responses
- Self-hosting
Every score we have
Each bar compares this model with the best result any model on this page has on the same test.
- LMArena text arena (human votes)1,474.6Arena score 1474.6 (95% CI 1468.4 to 1480.9); 12,793 votes; entry "deepseek-v4.1-flash-max"; position 38 of 416 on the full board
- LMArena creative writing1,440Arena score 1440.0 (95% CI 1427.6 to 1452.4); 2,597 votes; entry "deepseek-v4.1-flash-max"; position 58 of 414 on the full board
- LMArena WebDev (coding)1,618.4Arena score 1618.4 (95% CI 1607.5 to 1629.3); 3,540 votes; entry "DeepSeek-V4.1-Flash (Max)"; position 24 of 144 on the full board
- LMArena vision1,260.3Arena score 1260.3 (95% CI 1249.4 to 1271.2); 3,559 votes; entry "deepseek-v4.1-flash-max"; position 46 of 165 on the full board
- LMArena Agent Arena0.027IPS score 0.0274 (95% CI 0.0238 to 0.0310); 241,716 sessions; entry "DeepSeek-V4.1-Flash (Max)"; position 19 of 52 on the full board
- Artificial Analysis Intelligence Index39.539.5 on the Artificial Analysis Intelligence Index v4.3.2 (AA shows 40), setting "DeepSeek V4.1 Flash (max)"
- Humanity's Last Exam (Artificial Analysis)39.2%39.2% on Humanity's Last Exam (AA, 2,158 text-only questions), setting "DeepSeek V4.1 Flash (max)"
- Terminal-Bench 4.0 (Artificial Analysis run)26.8%26.8% on Terminal-Bench 4.0 as run by AA (mini-swe-agent harness), setting "DeepSeek V4.1 Flash (max)"
- Long-context reasoning (AA-LCR)84.0%84.0% on AA-LCR (long-context reasoning), setting "DeepSeek V4.1 Flash (max)"
- MMMU-Pro, image understanding77.0%77.0% on MMMU-Pro (image understanding, AA), setting "DeepSeek V4.1 Flash (max)"
- AA-Omniscience, accuracy minus hallucination-5.3-5.3 on the AA-Omniscience Index (knowledge accuracy minus hallucination, range -100 to 100), setting "DeepSeek V4.1 Flash (max)"
- Output speed, tokens per second176 t/s176 output tokens per second (AA median), row "DeepSeek V4.1 Flash (max)"
- Cost per Artificial Analysis test task$0.27$0.27 to run one AA Intelligence Index task, row "DeepSeek V4.1 Flash (max)"
- ARC-AGI-2 puzzles72.9%72.92% on ARC-AGI-2 (semi-private set), best listed setting "DeepSeek V4.1 Flash (Max)", $0.13 per task
- ARC-AGI-3 puzzles0.8%0.85% on ARC-AGI-3 (semi-private, standard harness; "provider adapter" runs excluded), best listed setting "DeepSeek V4.1 Flash (Max)"
Compare with
- Claude Opus 5.5Anthropic's flagship for long-running agentic coding and knowledge work, and the model Anthropic suggests starting with for most jobs.
- Claude Sonnet 5.5Anthropic's mid-priced model, which Anthropic describes as its best mix of speed and intelligence.
- Claude Fable 5.1Anthropic's top tier for the hardest reasoning and long-horizon agent work, which Anthropic recommends when Opus 5.5 still falls short.
- Claude Haiku 5.5Anthropic's smallest and fastest current model, built for high-volume, latency-sensitive jobs like classification, extraction and routing.