Where it stands
DeepSeek confirmed on 17 October 2026 that DeepSeek-V4.1-Flash-1017 is live on its website, app and API under the name deepseek-flash. The docs list a 1M context, 384K max output, vision support and off-peak input at $0.15 per 1M tokens (cache miss). DeepSeek says it beats the previous version on benchmarks, speed and output length, but the post gives no figures.
DeepSeek releases V4.1 Flash 1017 on web, app and API
DeepSeek-V4.1-Flash-1017 is live on the DeepSeek website, app and API, the company announced on X on 17 October 2026. The earlier beta ID is no longer the way to reach it: the API docs list the model name as deepseek-flash, now served by DeepSeek-V4.1-Flash-1017, and the calling method is unchanged.
What DeepSeek says changed
DeepSeek says the release has stronger benchmark performance, faster task completion and shorter outputs than the previous version, which it calls a Pareto improvement in performance and cost. The post gives no benchmark figures, so those gains are DeepSeek's own description for now.
The company says the model is optimised for DeepSeek Harness, Tencent CodeBuddy, WorkBuddy and other tools, and recommends upgrading to DeepSeek Harness v0.2.1-rc.1 or later for the best experience. The API docs describe DeepSeek Harness as being in developer preview for agent harness developers.
Model name and legacy IDs
The docs say deepseek-flash is the name to use. The older names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted, but the models behind them have been retired. Requests to them are served by DeepSeek-V4.1-Flash-1017 and billed at the Flash price. The docs list the Flash model as supporting Vision, which the V4 Pro model (DeepSeek-V4-Pro-0813) does not.
Specs and pricing
The pricing page lists a 1M context length and a maximum output of 384K, with thinking mode on by default and non-thinking mode also supported. Flash supports JSON output, tool calls, the Responses API and the Anthropic API format. Its concurrency limit is 2,500, against 500 for V4 Pro.
Prices are per 1M tokens, and off-peak rates are half the peak rates. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays. Everything else, including weekends and those holidays, is off-peak.
- Input, cache hit: $0.003 off-peak, $0.006 peak
- Input, cache miss: $0.15 off-peak, $0.30 peak
- Output: $0.60 off-peak, $1.20 peak
For comparison, V4 Pro is listed at $0.66 (off-peak) and $1.32 (peak) for cache-miss input, and $1.98 and $3.96 for output.
A worked example
Take a request with 100,000 uncached input tokens and 10,000 output tokens. On Flash that costs about $0.021 off-peak (0.1 × $0.15 + 0.01 × $0.60) and $0.042 at peak. The same request on V4 Pro costs about $0.086 off-peak (0.1 × $0.66 + 0.01 × $1.98), roughly four times the Flash price. Cache hits would lower the input portion further.
Search and Fetch
Search and Fetch are separate APIs billed at $0.75 per 1,000 successful requests. Server-side tool calls made through DeepSeek Harness, Claude Code or Codex hit these APIs but are free for now. Per-request billing starts at 00:00 Beijing time on 1 November 2026.
What it means
This is DeepSeek's own release, not a leak or a beta sighting. Developers can call deepseek-flash today with the OpenAI-format base URL https://api.deepseek.com or the Anthropic-format URL https://api.deepseek.com/anthropic. The pricing page does not give a previous Flash price, so the size of the claimed cost improvement cannot be calculated from these figures.
DeepSeek V4.1 Flash API Beta: What We Know
The DeepSeek V4.1 Flash API beta is reportedly live, giving developers access to an intermediate build identified as `deepseek-v4.1-flash-expires-on-0910`. In a post dated September 8, 2026, @Conor_D_Dart says the model is already available to call through the API.
What the announcement says
The post describes DeepSeek V4.1 Flash as a build with a new architecture and native multimodality. “Native multimodality” generally refers to handling more than one type of input or output as part of the model’s design, but the announcement does not specify which modalities are supported or how they can be used through the API.
The reference to a new architecture is similarly broad. No technical details, performance results or comparison with another DeepSeek model are included, so the announcement establishes the direction of the beta rather than explaining how the underlying system works.
The model ID and beta availability
The model ID is the clearest practical detail for anyone looking for the beta: `deepseek-v4.1-flash-expires-on-0910`. The post presents this identifier as the one associated with the available API build.
The `expires-on-0910` wording suggests that the build may be intended for a short testing window, potentially before a broader V4.1 release. That interpretation is not a confirmed expiration policy or launch schedule, however. The announcement does not say what happens after the date in the identifier, when a full V4.1 release will arrive, or whether access is available to every API user.
For now, the supported takeaway is limited but useful: DeepSeek V4.1 Flash is being presented as an API beta with a distinct model ID, a new architecture and native multimodality. Developers would need to consult the relevant API documentation and account controls for access and usage details, which the announcement does not provide.
0 comments
No approved comments yet. You can start the conversation.
Leave a comment