DeepSeek just dropped V4-Pro-0813 with zero announcement. No blog post, no tweet. Just quietly flipped the API version. Architecture's unchanged — still 1.6T MoE / 49B active — but post-training clearly got reworked. Agent performance jumped. 1M context, 384K output, three reasoning tiers. Also added Anthropic protocol support so Codex and Claude Code work out of the box with no changes. Grok-4.6 just dropped, now this. Gonna run some benchmarks and post a comparison soon.
DeepSeek just dropped V4-Pro-0813 with zero announcement. No…