DeepSeek V4.1 Flash at a glance
| DeepSeek-V4.1-Flash-1017 | |
|---|---|
| Released | V4.1 Flash on September 10, 2026; the 1017 update on October 17, 2026 |
| API model name | deepseek-flash (older names deepseek-v4-flash and deepseek-v4-flash-vision-exp are served by it) |
| Input | Text and images |
| Context window | 1M tokens; DeepSeek recommends max_tokens of 256K or more |
| Weights | Open on Hugging Face, MIT license |
| Where it runs | DeepSeek API, website and app; default in DeepSeek Harness |
| Tuned for | DeepSeek Harness (v0.2.1-rc.1 or later), Tencent CodeBuddy and WorkBuddy, per DeepSeek |
DeepSeek V4.1 Flash price vs V4 Pro
DeepSeek API prices in US dollars per million tokens on November 6, 2026. Off-peak rates apply outside 01:00–04:00 and 06:00–10:00 UTC on weekdays and are half of peak; the 1017 update didn't change prices. DeepSeek's Search and Fetch APIs cost $0.75 per 1,000 calls each.
| Per 1M tokens | Flash, peak | Flash, off-peak | V4 Pro, peak | V4 Pro, off-peak |
|---|---|---|---|---|
| Input, cache hit | $0.006 | $0.003 | $0.044 | $0.022 |
| Input, cache miss | $0.30 | $0.15 | $1.32 | $0.66 |
| Output | $1.20 | $0.60 | $3.96 | $1.98 |
DeepSeek V4.1 Flash benchmarks (as reported by DeepSeek)
These are DeepSeek's own numbers from the model card, comparing the 1017 update with the September V4.1 Flash. DeepSeek ran Terminal-Bench 4.0 with the Claude Code harness and DeepSWE with Mini-SWE-V2. Vendors' benchmarks favor their own models; test on your own tasks before switching.
| Benchmark | V4.1 Flash 1017 | V4.1 Flash (Sept) |
|---|---|---|
| Terminal-Bench 4.0 | 42.3 | 31.2 |
| DeepSWE v1.1 | 77.0 | 74.2 |
| ProgramBench | 29.1 | 20.3 |
| NL2Repo-Bench | 68.2 | 65.4 |
| Automation-Bench v1.0.6 | 60.5 | 54.8 |
| Agents' Last Exam | 35.4 | 31.8 |
| HLE (no tools) | 42.6 | 36.8 |
V4.1 Flash vs V4 Pro: which to use
- V4 Pro (
deepseek-v4-pro, generally available since August 13, 2026) costs 4.4 times as much per input token and 3.3 times as much per output token, and takes text only; Flash adds image input. - Both have a 1M-token context and adjustable reasoning effort (low, high and max).
- DeepSeek's own coding-agent setups default to Flash: DeepSeek Harness, and its recommended Claude Code settings, which map Sonnet and Haiku to Flash and only Opus to V4 Pro.
- A practical split: Flash for everyday agent work and subagents, V4 Pro when a hard task keeps failing on Flash.
How to use DeepSeek V4.1 Flash
Through the API
Get an API key on platform.deepseek.com and call the model as
deepseek-flash; the API is OpenAI- and Anthropic-compatible.curl https://api.deepseek.com/chat/completions \ -H "Authorization: Bearer $DEEPSEEK_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model": "deepseek-flash", "messages": [{"role": "user", "content": "Hello"}]}'In a coding agent
Run it in DeepSeek Harness, where it's the default, or point Claude Code or Codex at DeepSeek; see DeepSeek Harness and Claude Code with DeepSeek.
Locally, from the open weights
Download DeepSeek-V4.1-Flash-1017 from Hugging Face. The repository ships a reference prompt encoder and inference instructions instead of a chat template; DeepSeek recommends temperature 1.0 and top_p 1.0.
A cheap, capable model, working through your backlog
At Flash prices it's affordable to keep an agent busy all day. Crewmark turns a DeepSeek Harness, Claude Code, Codex or OpenCode session into an AI employee with a role, a goal and working hours, and wakes it on a heartbeat to take the next issue from a board.
Pick the agent and the model
DeepSeek Harness runs V4.1 Flash by default; Claude Code and Codex can use it through DeepSeek's endpoint.
Give it a role and a goal
Read what an AI employee is and how it differs from a superagent.
Let the heartbeat start the work
During working hours it wakes every 15, 30 or 60 minutes and takes the next issue; see how to make Claude Code proactive.
DeepSeek V4.1 Flash FAQ
What is DeepSeek V4.1 Flash 1017?
The October 17, 2026 update of DeepSeek V4.1 Flash, the smallest model in DeepSeek's newest architecture. It accepts text and images, has a 1M-token context, and is served as deepseek-flash.
How much does DeepSeek V4.1 Flash cost?
On the API, $0.30 per million input tokens on a cache miss ($0.006 on a hit) and $1.20 per million output tokens at peak; off-peak prices are half that.
What is the context window of DeepSeek V4.1 Flash?
1M tokens, with DeepSeek recommending max_tokens of at least 256K.
Is DeepSeek V4.1 Flash open source?
Its weights are open on Hugging Face under the MIT license, with a reference prompt encoder and inference instructions.
DeepSeek V4.1 Flash vs V4 Pro: what's the difference?
V4 Pro costs 4.4 times as much per input token and 3.3 times per output token, and is text-only; Flash accepts images and is the default for DeepSeek's coding-agent setups. Both have a 1M context.
What changed from V4.1 Flash to 1017?
DeepSeek reports higher scores on agent and coding benchmarks, shorter outputs and faster task completion, at the same price and model name.
Can I use DeepSeek V4.1 Flash in Claude Code?
Yes. DeepSeek documents environment variables that point Claude Code at its Anthropic-compatible endpoint with deepseek-flash; see Claude Code with DeepSeek.
Checked against DeepSeek's API docs (change log, news, pricing) and the Hugging Face model card on November 6, 2026.