April 23, 2026
Docs or it didn’t drop
DeepSeek v4
DeepSeek v4: Docs drop, “weights” whisper, community goes full soap opera
TLDR: DeepSeek updated its API with OpenAI/Anthropic‑compatible models (v4‑pro and v4‑flash) and plans to retire older names in 2026. Commenters are split between “it’s just docs,” “the weights are live on Hugging Face,” and big performance claims—equal parts excitement and exhaustion over an easy drop‑in replacement.
DeepSeek quietly updated its API docs and the internet did what it does best: meltdown in stereo. The official line is simple enough—new models called deepseek-v4-pro and deepseek-v4-flash, plug‑and‑play with OpenAI and Anthropic tools (just change the website address), with a “thinking mode” you can switch on and off. Old names deepseek-chat and deepseek-reasoner are getting retired on 2026/07/24. There’s even a copy‑and‑paste code snippet and a switch for streaming replies. But the comments? Absolute chaos.
One side is yelling, “It’s just docs, not a real release,” as [luyu_wu] insists. Another camp is flexing receipts: [seanobannon] dropped a link to the weights on Hugging Face, implying you can actually download the model files. Then [nthypes] waltzed in with a PDF and a bold claim: this thing is “frontier level,” beating a top-tier rival while costing less. Cue the oohs, aahs, and side-eyes.
Meanwhile, someone asked “Merge?” like they tossed a mystery grenade into the thread, and another commenter begged for a support group because AI news is dropping faster than caffeine hits. The vibe: Is it out or not? Hype vs. receipts. Drop-in replacement joy vs. naming confusion fatigue. And the meme of the day? “We need AI to keep up with AI.”
Key Points
- •DeepSeek’s API supports OpenAI and Anthropic API formats via configurable base URLs.
- •Available models include deepseek-v4-flash and deepseek-v4-pro.
- •Legacy models deepseek-chat and deepseek-reasoner will be deprecated on 2026/07/24 and map to deepseek-v4-flash non-thinking and thinking modes, respectively.
- •A sample curl request shows how to call the Chat Completions endpoint with an API key, enable thinking, and set reasoning_effort.
- •Streaming responses are supported by setting the stream parameter to true, and Anthropic-format examples are referenced externally.