DeepSeek V4 Flash
deepseek/deepseek-v4-flash- DeepSeek
- Family · DeepSeek V4 Flash
- Released 31 Jul 2026
- Open weights
- Text → Text
- Reasoning · adjustable
- Tools · JSON
Spec & list price
Context1M
Max output944K
Input / 1M$0.09
Output / 1M$0.182× input
Scores · standing
Intelligence34.3#51/489
Coding69.1#28/196
Usage
Tokens / day323.3B▼ 8.4% 7d
Usage rank#19
Requests / day25.0M
Tool calls / req0.10
Per request · 30 days
Avg input7.7K0.5× median
Avg output5280.8× median
Cache hit62%0.9× median
Reasoning share53%about median
Tool-call failures2.4%1.3× median
Consumption
Peak 549.4B 13 OctLow 204.4B 1 NovRank today #19
News & trends
12 mentions this week ▲ 71%this model · videos, posts, articles · 30 days
Trending · 14 days
- 【実測】新deepseek-v4-flashに議事録要約を任せたら3秒→5秒80 views
- Recherche web : Gemini donne 3 puces, DeepSeek V4 Flash sans outil | PREMIER TP 1343 views
- My voice agent stack: Deepgram + DeepSeek V4 Flash + Inworld, ~1s to first token. What would you change?20 upvotes
- DeepSeek V4 Flash built a walkable A320 cabin in 51 minutes for $0.80479 upvotes
News · latest from DeepSeek V4 Flash
- Tell HN: Mistrap Pro includes DeepSeek v4.1 Flash 1017docs.mistral.ai · 1 Nov
- Show HN: DeepSeek v4.1 Flash-1017 Safety Evaluationtwitter.com · 26 Oct
- DeepSeek-V4 Flash 0731 vs GPT-5.6 Luna on DeepSWE: Cost and CodingTogether AI · 6 Aug
Benchmarks
bar = share of the top score · Benchmark data: Artificial AnalysisSpeed · median across 16 providers, latest day
Output speed#193/33458 tok/s1.1× slower than median
Fastest provider175 tok/sbest of 16 providers
First token#1580.98 sabout median
Best value in its score band
Value
Intelligence Index
Blended price per 1M tokens (log)
Family progress
Intelligence Index
- Family releases45.1
Technical details
- Reasoning
- on · high
- Tokenizer
- DeepSeek
- License
- MIT
- Parameters
- 291B
- Moderation
- none
- Downloads · likes
- 1.1M · 2.4K
Reasoning effort
- xhigh
- high · default
- none
Request parameters 22
- tools
- tool_choice
- response_format
- structured_outputs
- reasoning
- reasoning_effort
- include_reasoning
- temperature
- top_p
- top_k
+12 more
- min_p
- max_tokens
- max_completion_tokens
- stop
- seed
- frequency_penalty
- logit_bias
- logprobs
- presence_penalty
- repetition_penalty
- top_a
- top_logprobs
When to use
Best for
- Low-cost coding assistants and code review
- Research and analysis over very long documents
- Tool-calling agents with structured output
- Self-hosted deployments (open weights)
Think twice for
- Image or screenshot understanding (text only)
- Browser-driving agents
- Long multi-step autonomous agent runs
Research papers
- FlatCake: A Flat KVCache Storage Architecture for LLM Serving28 Oct
- TrawMem: Reframing Agent Memory as Task-Oriented Role-Adaptive Workspaces27 Oct
- Learning to Formalize: Bridging Open-Ended Workloads and Fast Decision Models27 Oct
- SEMA-Auto: Sighted and Efficient Multi-Turn Attack with Autonomy26 Oct
- Uncovering the Turing Test's Latent Signal: Indistinguishability as a Reward26 Oct
- When Agents Fail: Model or Harness--Who's to Blame?23 Oct
- HyScale: Scaling Long-Context Training for Learned Sparse Attention23 Oct
- SWE-VERIFIER: Selective Verification for Cost-Efficient Coding Agents21 Oct
Social
Most watched and discussed · 90 days
- Recherche web : Gemini donne 3 puces, DeepSeek V4 Flash sans outil | PREMIER TP 1Mais que fait l’IA ? · 343 views · 27 Oct
- 【実測】新deepseek-v4-flashに議事録要約を任せたら3秒→5秒今日のAIニュース · 80 views · 28 Oct
- @Uncrashed_FPV_Sim Deepseek v4-flash 1017 гоночный автопилот. #uncrashedLuckyAnon · 57 views · 30 Oct
- Mô hình DeepSeek V4 Flash chạy trên chỉ tám card L4 #Shortshọc AI · 14 views · 31 Oct
- Ray + vLLM พื้นฐาน และการ serve DeepSeek-V4-Flash | Ray Summit 2026 Recap EP1Supachai Jaturaprom (ตั้ม) · 6 views · 29 Oct
- DeepSeek V4 Flash Cyber Rilis: AI Logika Rumit & Coding Secepat Kilat #ShortsArijal Meutuwah · 2 views · 31 Oct
Same family
DeepSeek V4 Flash- 39.5
- 45.1
- 24.4
Alternatives
other vendors · similar capability- 38.1
- 37.9
- 37.3
- 33.7
Using DeepSeek V4 Flash at volume?
Sign in to ask for a volume discount. You will continue in the OSMEXO dashboard.