Independent guide · not affiliated with, sponsored by, or endorsed by DeepSeek (Hangzhou DeepSeek Artificial Intelligence Co., Ltd.) · Official site: deepseek.com · About this site
Which DeepSeek models you can use today, what they cost, which names are retired, and every model release since 2024, dated from DeepSeek's own change log.
Latest DeepSeek model
API name deepseek-flash · 1M context · up to 384K output · understands images · $0.15 input / $0.60 output per 1M tokens off-peak.
Latest change: Full application API suite for deepseek-flash (November 13, 2026).
| Model | API name | Context / max output | Images | Off-peak price per 1M (input / output) | Best for |
|---|---|---|---|---|---|
| DeepSeek-V4.1-Flash-1017 | deepseek-flash | 1M / 384K | Yes | $0.15 / $0.60 | Default for chat, agents, tool use, RAG and vision — fast and cheapest |
| DeepSeek-V4-Pro-0813 | deepseek-v4-pro | 1M / 384K | No | $0.66 / $1.98 | Hardest reasoning, long multi-step coding and math |
Peak hours (01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays) cost double. Cache hits are far cheaper. Full rates and a cost calculator are on our DeepSeek pricing page; how to call the models is in the API guide.
deepseek-v4-flash, deepseek-v4-flash-vision-exp
Still accepted, temporarily routed to DeepSeek-V4.1-Flash and billed at the Flash price (since 10 September 2026).
Use instead: deepseek-flash
deepseek-chat, deepseek-reasoner
Scheduled for discontinuation on 24 July 2026 and no longer in the official model list.
Use instead: deepseek-flash (thinking off for the old chat behaviour, on for reasoner)
Every entry in DeepSeek's official API change log, newest first. The change log starts in May 2024.
Web search, file input, URL reading and model safety become available for deepseek-flash in the Messages, Responses and Chat Completions APIs. DeepSeek also adds standalone Search and Fetch, OCR, speech-to-text and text-to-speech APIs, plus an open-source demo chat client.
API: deepseek-flash.
The model behind deepseek-flash is upgraded to DeepSeek-V4.1-Flash-1017. The way you call it and the price stay the same. DeepSeek also opens its own search API.
API: deepseek-flash.
DeepSeek-V4.1-Flash, with native image understanding, ships as deepseek-flash and Flash prices come down (from 04:00 UTC). V4-Flash and V4-Flash-Vision-Exp are retired; their names are routed to V4.1-Flash. V4-Pro stays available with unchanged billing.
API: deepseek-flash. Our coverage
An experimental multimodal version of V4-Flash, under its own model name.
API: deepseek-v4-flash-vision-exp.
V4-Pro reaches general availability with stronger agent skills, native Responses API support for Codex, and low / high / max thinking effort. DeepSeek announces peak and off-peak pricing, with off-peak at half the peak price, taking effect at 16:00 UTC on 16 August 2026.
API: deepseek-v4-pro. Our coverage
A re-post-trained V4-Flash (same architecture and size as the preview) goes into public beta with stronger agent skills and Responses API support for Codex. V4-Pro and the app are unchanged.
API: deepseek-v4-flash. Our coverage
The V4 generation arrives on the existing base URL, with OpenAI and Anthropic compatible interfaces. The old deepseek-chat and deepseek-reasoner names are scheduled to stop on 24 July 2026 and until then map to V4-Flash without and with thinking.
API: deepseek-v4-pro, deepseek-v4-flash. Our coverage
deepseek-chat and deepseek-reasoner move to DeepSeek-V3.2. V3.2-Speciale runs on a temporary endpoint until 15 December 2025, without tool calls.
API: deepseek-chat, deepseek-reasoner.
Both API names move to the experimental V3.2-Exp.
API: deepseek-chat, deepseek-reasoner.
An update to V3.1 with more consistent language and better coding and search agents.
API: deepseek-chat, deepseek-reasoner.
V3.1 serves both names: deepseek-chat without thinking, deepseek-reasoner with thinking.
API: deepseek-chat, deepseek-reasoner. Our coverage
An upgraded R1 with large gains on maths, science and coding tests; it can use more tokens per answer than the first R1.
API: deepseek-reasoner. Our coverage
An upgraded V3 with better reasoning, front-end code, Chinese writing and function calling.
API: deepseek-chat.
R1 launches as deepseek-reasoner and makes DeepSeek known worldwide.
API: deepseek-reasoner. Our coverage
deepseek-chat moves to V3. The API itself does not change.
API: deepseek-chat.
A final V2.5 update with better maths and coding, plus improved file upload and web page summaries.
API: deepseek-chat.
V2 Chat and Coder-V2 become one model, reachable under both names.
API: deepseek-chat, deepseek-coder.
Repeated prompt prefixes are cached automatically and billed at a much lower cache-hit price.
An update to the coding model. The next day the API gains JSON mode, function calling and fill-in-the-middle completion.
API: deepseek-coder.
An update to V2 Chat with better coding, maths and role-play.
API: deepseek-chat.
Coder-V2 arrives on the API.
API: deepseek-coder.
The first entry in DeepSeek's API change log: V2 Chat with better instruction following and JSON output.
API: deepseek-chat.
DeepSeek-V4.1-Flash-1017, served in the API as deepseek-flash. It replaced the first V4.1-Flash on 17 October 2026. For the hardest reasoning DeepSeek still offers DeepSeek-V4-Pro-0813 as deepseek-v4-pro.
Start with deepseek-flash: it is the cheapest, the fastest, understands images and handles most chat, agent and coding work. Switch to deepseek-v4-pro only for the hardest multi-step reasoning, coding and maths, where it costs about 3.3 times as much per output token.
No. As of November 19, 2026, DeepSeek's official change log has no V5 release. DeepSeek has said a V4.1-Pro will follow V4.1-Flash, without giving a date.
They were the API names for the V3 and R1 generations. DeepSeek scheduled them to stop on 24 July 2026; use deepseek-flash with thinking off for the old chat behaviour, or with thinking on for the old reasoner behaviour.
From DeepSeek's official API change log and release notes. Dates are DeepSeek's own. We check the change log for new entries and update this page when DeepSeek ships something.