DeepSeek V4 Pro 0813 goes GA: 1M context at $0.44/M input
The four-month preview ends with agent-grade coding scores and OpenAI- and Anthropic-format endpoints — plus Codex reaches Linux and Claude Code fixes Slack.
Copy markdown
The flagship leaves preview
DeepSeek's V4 Pro 0813 is now generally available — a mixture-of-experts model with 1.6T total parameters, 49B active per token, and three modes from non-thinking to maximum reasoning. Vendor benchmarks put it at 80.6% on SWE-bench Verified and 93.5% on LiveCodeBench, though independent confirmation is still pending.
The price is the real headline
Input runs $0.435 per million tokens and output $0.87, with cache hits near $0.0036/M — frontier-class coding at a fraction of Opus or GPT rates. The context window is 1M tokens with up to 384K tokens of output.
Drop it into your existing stack
V4 Pro speaks both OpenAI ChatCompletions and Anthropic Messages formats, with tool calling and JSON output, and routes through three OpenRouter providers (Balanced, Nitro, Exacto). You can repoint a Claude or GPT client at it without rewriting your code.
But the 0813 weights aren't out
Only the April preview weights sit on Hugging Face (1.4M+ downloads last month); DeepSeek has set no timeline for publishing the 0813 build. For now it's API-only — no local self-host — and the headline scores remain vendor-reported.
Elsewhere: Codex lands on Linux
OpenAI shipped its ChatGPT + Codex desktop app to Linux in preview — Ubuntu 24.04/26.04, Debian 13, Fedora 43/44 — closing its last platform gap about a month after Anthropic's Claude desktop. Codex now runs natively outside the browser on every major OS.
Elsewhere: Claude Code 2.1.231 unbreaks Slack
The latest Claude Code point release fixes a critical OAuth sign-in bug that was blocking connections to services like Slack. If your connectors dropped after 2.1.229, update to restore them.