DeepSeek's V4-Pro-0813 quietly outperforms Opus 4.8 in coding and agents

DeepSeek has silently upgraded its API to DeepSeek-V4-Pro-0813, released August 13, 2026, without changing the model ID. That means developers who were already using the endpoint automatically get the new capabilities with zero migration friction. The stealth release is notable because V4-Pro-0813 improves significantly on agentic tasks—AI systems that plan and execute multi-step actions—and on coding benchmarks, where it now outperforms Anthropic's Opus 4.8 in several head-to-head evaluations. This puts DeepSeek back in the top tier of coding and agent-oriented models, at a fraction of the usual cost.

The model retains its 1 million-token context window, making it suitable for long documents, large codebases, and complex multi-turn agent workflows. Pricing remains brutally competitive: cache hit input costs just 0.025 yuan per million tokens (≈$0.0035), while output runs 6 yuan per million tokens (≈$0.85). That's roughly 10x cheaper than leading US models for comparable performance. The quiet launch strategy—no announcement, no fanfare—suggests DeepSeek is confident the benchmarks and price point will speak for themselves. For AI engineers, this means they can build agentic systems and code-assist tools with frontier-level quality while slashing inference budgets, and all they need to do is keep calling the same endpoint.