{"slug": "deepseek-open-sources-coding-agent-harness-before-v4-pro-api-price-hike", "title": "DeepSeek open-sources coding-agent harness before V4-Pro API price hike", "summary": "DeepSeek released DeepSeek Harness v0.1, an open-source runtime for building coding agents, on August 13th under the MIT license, alongside the general-availability version of DeepSeek-V4-Pro (DeepSeek-V4-Pro-0813) and a price increase scheduled for August 16th. The harness, built on the Cordis plugin framework, allows developers to replace models, tools, and other components, and supports multiple providers including Anthropic, OpenAI, and Amazon Bedrock. The repository had about 94,000 stars and 8,600 forks on GitHub by August 14th, indicating significant launch attention.", "body_md": "Liang Wenfeng's [DeepSeek (@deepseek_ai)](https://x.com/deepseek_ai) released an open-source runtime for building coding agents on August 13th, pushing beyond model APIs and into the developer tooling controlled by Anthropic's Claude Code and OpenAI's Codex. The release arrived alongside the general-availability version of DeepSeek-V4-Pro and a price increase scheduled for August 16th.\n\n[DeepSeek's official changelog](https://api-docs.deepseek.com/updates/) confirms that V4-Pro reached DeepSeek's app, web interface and API on August 13th. [VentureBeat reported](https://venturebeat.com/technology/deepseek-harness-launches-as-open-source-rival-to-claude-code-alongside-v4-pro-on-api-with-higher-prices) that DeepSeek Harness v0.1 entered developer preview the same day under the MIT license.\n\nThe paired launch marks a change in scope for [Liang](https://www.investing.com/news/economy-news/deepseek-rushes-to-launch-new-ai-model-as-china-goes-all-in-3888553), the engineer and quantitative-finance founder who built the High-Flyer hedge fund before establishing DeepSeek in 2023. DeepSeek initially concentrated its resources on model research rather than elaborate consumer applications. Harness puts DeepSeek directly into the runtime layer that determines how a model reads files, invokes tools, executes commands and maintains work across multiple steps.\n\n### An open runtime around the model\n\n[DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness), also called `dsh`\n\n, is built on the Cordis plugin framework. Its core design rule is that practically every component can be replaced: models, tools, skills, sessions, sandboxes, storage, agent loops, orchestration and user interfaces.\n\nDevelopers can start a local web interface with:\n\n```\nnpx @deepseek-ai/dsh web\n```\n\nThe standard agent mode can inspect and edit files, run shell commands, search local files and the web, maintain a plan, invoke reusable skills, delegate tasks to subagents and apply approval policies. DeepSeek also provides a Python SDK and modes tailored to tool composition, stripped-down benchmarking and custom agent creation.\n\nThe model itself is another plugin. DeepSeek's [provider documentation](https://deepseek-harness.github.io/deepseek-harness/en/guide/providers) supports Anthropic, OpenAI, Codex, Amazon Bedrock, Google Vertex AI, Microsoft Azure and custom endpoints alongside DeepSeek's API. That makes Harness useful as a model-agnostic framework, even though DeepSeek gains a natural distribution channel for V4-Pro and V4-Flash.\n\nHarness currently occupies a different product tier from [Claude Code](https://docs.anthropic.com/en/docs/claude-code/getting-started) and [Codex](https://developers.openai.com/). Anthropic and OpenAI have built managed coding products across terminals, development environments, cloud tasks and repository workflows. DeepSeek is shipping a composable runtime and local interface that developers can inspect, fork and rebuild.\n\nDeepSeek also warns that Harness remains a developer preview and will receive compatibility-breaking changes. The MIT license removes a legal barrier to commercial use, while the unstable interfaces raise the engineering cost of deploying it in production today.\n\nInterest has moved faster than the software's maturity. GitHub displayed about 94,000 stars and 8,600 forks for the repository on August 14th. Those figures measure launch attention rather than active installations or production adoption.\n\n### V4-Pro reaches general availability\n\nDeepSeek first introduced the V4 family in preview on April 24th. V4-Pro has 1.6 trillion total parameters, with 49 billion active for each token, according to DeepSeek. The August 13th build, identified as DeepSeek-V4-Pro-0813, is positioned around coding and long-running agent tasks.\n\nDeepSeek says the general-availability release scored 87.9 on Terminal Bench 2.1, 62.7 on DeepSWE and 60.0 on Humanity's Last Exam with tools. These are vendor-reported results, and several comparisons include DeepSeek's internal test sets. The release also adds three reasoning-effort settings - low, high and max - plus native support for the OpenAI Responses API format and configuration for Codex.\n\nHarness matters to those benchmark claims. In its July 31st V4-Flash release note, DeepSeek said its public code-agent evaluations used the then-unreleased Harness in minimal mode, configured with only a persistent shell and a file-editing tool. Publishing Harness gives developers a path to inspect and reproduce more of the runtime around the model, where tool definitions, prompts and execution loops can materially affect agent scores.\n\n### DeepSeek resets API pricing\n\nDeepSeek will replace its flat international API rates with peak and off-peak pricing at 16:00 UTC, or 12 p.m. Eastern, on August 16th. Peak periods will run from 01:00 to 04:00 UTC and from 06:00 to 10:00 UTC. All other hours will use the off-peak rate.\n\nFor V4-Pro, the change is an increase at every level:\n\n| Per 1 million tokens |\nCurrent flat rate |\nNew off-peak rate |\nNew peak rate |\n| Cached input |\n$0.003625 |\n$0.022 |\n$0.044 |\n| Uncached input |\n$0.435 |\n$0.66 |\n$1.32 |\n| Output |\n$0.87 |\n$1.98 |\n$3.96 |\n\nEven the discounted rate raises cached-input pricing by about 507%, uncached-input pricing by 52% and output pricing by 128%. During peak hours, cached input costs roughly 12 times the current rate, while output rises by 355%.\n\nThe combination gives DeepSeek a broader developer strategy. Harness can spread as open infrastructure across multiple model providers, while V4-Pro becomes the default commercial endpoint sitting beside it. Developers gain control over the agent runtime and lose part of the pricing advantage that originally made DeepSeek attractive for cache-heavy, long-running workloads.\n\nThe release also puts DeepSeek's agent benchmark results and API economics in the same frame. Long sessions, repository analysis and subagent workflows consume repeated input and output tokens. Harness is designed to make those workloads easier to build just as DeepSeek is charging substantially more to run them on V4-Pro.", "url": "https://wpnews.pro/news/deepseek-open-sources-coding-agent-harness-before-v4-pro-api-price-hike", "canonical_source": "https://runtimewire.com/article/deepseek-open-sources-agent-harness-v4-pro-api-price-hike", "published_at": "2026-08-15 01:15:04+00:00", "updated_at": "2026-08-15 01:40:44.972454+00:00", "lang": "en", "topics": ["developer-tools", "ai-tools", "ai-products", "artificial-intelligence"], "entities": ["DeepSeek", "Liang Wenfeng", "DeepSeek Harness", "DeepSeek-V4-Pro", "Cordis", "Anthropic", "OpenAI", "GitHub"], "alternates": {"html": "https://wpnews.pro/news/deepseek-open-sources-coding-agent-harness-before-v4-pro-api-price-hike", "markdown": "https://wpnews.pro/news/deepseek-open-sources-coding-agent-harness-before-v4-pro-api-price-hike.md", "text": "https://wpnews.pro/news/deepseek-open-sources-coding-agent-harness-before-v4-pro-api-price-hike.txt", "jsonld": "https://wpnews.pro/news/deepseek-open-sources-coding-agent-harness-before-v4-pro-api-price-hike.jsonld"}}