Overview
<cite index="4-3,4-4">DeepSeek has officially launched DeepSeek V4-Pro, moving its flagship large language model (LLM) out of preview and into general release. The new build, designated V4-Pro-0813, supersedes the preview version that had been available since April.</cite> <cite index="4-1">The model is now available through app, web, and application programming interface (API) platforms.</cite>
Model Architecture and Capabilities
<cite index="15-5,15-6">DeepSeek V4-Pro is a large-scale Mixture-of-Experts model with 1.6 trillion total parameters and 49 billion activated parameters, supporting a one-million-token context window. It is designed for advanced reasoning, coding, and long-horizon agent workflows, with strong performance across knowledge, math, and software engineering benchmarks.</cite> <cite index="1-7">Users can choose between thinking and non-thinking modes, and the model can generate outputs of up to 384,000 tokens.</cite>
<cite index="4-5">The release is centered on agentic capabilities—the growing category of AI tasks in which a model uses external tools, writes and executes code, and carries out multi-step workflows with little or no human supervision.</cite> <cite index="1-5">According to DeepSeek's own benchmark findings, the model achieved results on various agent-focused tests, including Terminal Bench 2.1 (87.9), DeepSWE (62.7), and NL2Repo (61.5).</cite> <cite index="7-7">No independent evaluator has replicated those numbers.</cite>
<cite index="17-9">Both V4-Pro and V4-Flash now feature flexible reasoning capabilities at low, high, and max settings, along with thinking modes that use chain-of-thought (CoT) reasoning to improve answer accuracy.</cite> <cite index="1-8">The V4-Pro API has also been upgraded to natively support the OpenAI Responses API format and has Codex integration capabilities.</cite>
Pricing Changes
<cite index="3-1">DeepSeek said it will raise API prices for its V4-Pro and V4-Flash models, with increases ranging from 50% to 1,100% depending on the model, token type, and time of use.</cite> <cite index="13-1,13-2">Starting August 16, V4-Pro will cost $3.96 for one million output tokens at peak hours—more than four times the previous rate of $0.87—and $1.98 during off-peak hours.</cite> <cite index="9-8,9-9">The largest increase lands on cache-hit input for V4-Pro: during peak hours, the price rises from RMB 0.025 to RMB 0.30 per million tokens, a twelve-fold increase.</cite>
<cite index="13-9">The V4-Flash model will cost $1.32 for one million output tokens during peak hours (up from $0.28) and $0.66 during off-peak.</cite> <cite index="14-5">DeepSeek stated it is revising pricing "to allocate resources more reasonably."</cite> <cite index="9-12,9-13">Peak hours are defined as 09:00–12:00 and 14:00–18:00 Beijing time; all other hours are off-peak, priced at half the peak rate.</cite>
<cite index="13-12,13-13,13-14">Even so, DeepSeek's rates remain below those of some rivals: Kimi K3 from Chinese company Moonshot costs $15 per million output tokens, while OpenAI's GPT-5.6 Sol costs $30 per million output tokens.</cite>
DeepSeek Harness
<cite index="23-2">Alongside the model release, DeepSeek published DeepSeek Harness, an MIT-licensed agent harness whose architecture treats every component—models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the UI—as swappable plugins.</cite> <cite index="19-2">The Harness v0.1, now available as open-source software under the MIT license, is an agent runtime framework designed to transform DeepSeek's V4 series models into autonomous coding agents capable of multi-step workflows and tool use.</cite> <cite index="6-17">The open-source agent software is pitched as an alternative to OpenAI's Codex and Anthropic's Claude Code.</cite> <cite index="22-8">Within hours of publication, the repository picked up more than 33,000 GitHub stars.</cite>
Corporate Context
<cite index="30-5">DeepSeek closed its first external funding round in late May or early June 2026, pulling in over 50 billion yuan (approximately $7.4 billion) at a post-money valuation above $50 billion.</cite> <cite index="32-13">Investors reportedly included Chinese technology companies Tencent, JD.com, battery maker CATL, and video game publisher NetEase.</cite> <cite index="35-2,35-3">The company has since quietly restarted a second funding round, targeting 50 billion yuan at a pre-money valuation of approximately 500 billion yuan (approximately $74.1 billion), up roughly 43% from the first round.</cite>
<cite index="6-4">On independent benchmarks, V4-Pro scores higher on agent evaluations than its predecessor but still trails top models such as Claude Opus 5 in overall rankings.</cite> <cite index="24-2,24-3,24-4,24-5">For enterprise developers, Harness may prove the more consequential part of the announcement: models can increasingly be swapped behind standardized interfaces, but the harness that controls how an agent reasons, invokes tools, edits software, and persists across a workflow is much harder to replace—a layer DeepSeek is now competing for directly.</cite>