DeepSeek Harness launches as open-source agent framework alongside V4-Pro with higher API prices

DeepSeek launched DeepSeek Harness v0.1, an MIT-licensed open-source agent framework, alongside the official V4-Pro 0813 model with OpenAI Responses API support and new peak/off-peak API pricing. The move signals a two-layer strategy: giving developers open tooling to assemble agent workflows while raising API costs and nudging high-volume users toward self-hosting or open weights.

A modular agent harness built on Cordis

DeepSeek Harness is built on the Cordis framework with a guiding principle that everything is a plugin. Models, tools, skills, sessions, sandboxes, filesystems, loops, orchestration, and user interfaces can all be swapped or extended. Source This makes it a model-agnostic alternative to the agent infrastructure underlying Claude Code and OpenAI Codex, though it is not yet a full replacement for either product's broader developer experience.

The harness can inspect repositories, edit files, execute shell commands, search files and the web, maintain plans, invoke skills, delegate work to subagents, and enforce approval policies. It is available under the MIT license and can be launched from npm with npx @deepseek-ai/dsh web. Source The repository explicitly warns of compatibility-breaking changes, so enterprise teams should treat it as a developer preview rather than a stable production platform.

V4-Pro 0813: agent-tuned model with standardized interfaces

The official V4-Pro 0813 release transitions the model from preview to general availability with significantly enhanced agent capabilities. It adds native OpenAI Responses API support and Codex optimization, lowering integration work for applications already built around those interfaces. Source The model also introduces configurable reasoning effort with three levels: Non-think for fast routine tasks, Think High for complex problem-solving, and Think Max for maximum reasoning on difficult problems.

DeepSeek reports V4-Pro-0813 scores of 87.9 on Terminal Bench 2.1 and 74.1 on Toolathlon-Verified, though these are company-reported figures and some results were obtained using DeepSeek Harness in minimal mode. Source That means the agent benchmarks measure the model operating inside the harness environment, not the model alone.

API pricing shifts to peak and off-peak rates

Beginning August 16 at 16:00 UTC, DeepSeek abandons flat API pricing for a peak/off-peak model. Peak hours are 01:00-04:00 UTC and 06:00-10:00 UTC. For V4-Pro, off-peak cache-miss input is $0.66 per million tokens and output is $1.98 per million tokens. Peak rates double to $1.32 input and $3.96 output. Source Cache-hit pricing also increases substantially, from $0.003625 per million tokens to $0.022 off-peak and $0.044 peak.

For a simple workload of one million cache-miss input tokens plus one million output tokens, V4-Pro currently costs $1.305. The same token mix will cost $2.64 off-peak (roughly double) or $5.28 during peak hours (more than four times the current price). Source The increases range from 50% to over 1,100% depending on model, token category, and time of use.

What this means for AI builders

DeepSeek is now competing on two fronts simultaneously. With Harness, it offers an open, modular framework for building agent workflows that can use any model, not just DeepSeek's own. With the pricing changes, it is making its hosted API significantly more expensive, especially during peak hours. For teams already running DeepSeek in production, workload scheduling and caching behavior become critical cost factors. The option to run open weights on self-hosted infrastructure becomes more attractive as API prices rise.

The harness layer may ultimately be the more consequential part of this release. 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 harder to replace. DeepSeek is now competing for that layer too, with an MIT-licensed framework that gives developers unusual flexibility at the cost of production maturity.