<cite index="23-1">On August 13, 2026, DeepSeek moved DeepSeek-V4-Pro from preview to general availability across its app, web experience, and API.</cite> The production release is designated V4 Pro 0813 and marks the end of a nearly four-month preview period. <cite index="24-2">The general availability version focuses on agent capabilities — tasks where AI systems use tools, execute code, and complete multi-step workflows without human intervention.</cite>
<cite index="24-2">Benchmark results released by DeepSeek showed the model scored 87.9 on Terminal Bench 2.1, 62.7 on DeepSWE, and 61.5 on NL2Repo, among other agent-focused tests. The model is capable of handling a context window of up to 1 million tokens and can produce outputs as long as 384,000 tokens, with the option to run in either thinking or non-thinking mode.</cite> <cite index="23-2">The official guidance positions low effort for straightforward work, high for everyday agent tasks, and max for the hardest scenarios.</cite>
<cite index="21-2">A price increase for the V4 model family is set to take effect at 16:00 UTC on August 16. DeepSeek is also introducing peak and off-peak billing, with off-peak rates at half the peak-hour price. V4-Pro output tokens rising to $3.96 per million at peak hours from the current flat rate of $0.87 per million.</cite> Even at peak rates, DeepSeek remains substantially cheaper than Anthropic's Fable 5 ($50 per million output tokens).
For production teams, V4 Pro GA signals a stable checkpoint after months of preview iteration. The pricing change reflects DeepSeek's shift from aggressive undercutting to sustainable margin. Key decision: whether the agent-benchmark gains justify migration from V4 Flash 0731 (which achieved parity or better on many coding tasks at 35x lower cost), or whether the tiered reasoning effort levels unlock new use cases at the premium price tier.