--- title: "Deepseek V4 Pro: Open-Sourcing Agents, Raising API Prices" description: "Deepseek's V4 Pro launch combines a significant performance boost (Terminal Bench 2.1 score from 72.1 to 87.9) with open-sourcing its agent software and raising API prices, signaling a new ecosystem-driven strategy." url: "https://www.thesocialalgorithm.work/blog/deepseeks-v4-pro-launch-agent-open-sourcing-price-hike" source: "generated from the same data as the HTML page" --- # Deepseek V4 Pro: Open-Sourced Agents and API Price Hikes Signal New AI Strategy > **The short answer** > > Deepseek has launched its V4 Pro model, open-sourced its Harness v0.1 agent software, and increased API prices. This strategic trifecta aims to build an ecosystem around its open-source agent framework while charging a premium for the powerful V4 Pro API, evidenced by a Terminal Bench 2.1 score jump from 72.1 to 87.9. > **Key facts** > > - Deepseek launched V4 Pro (build V4-Pro-0813), its flagship model. > > - Deepseek open-sourced its Harness v0.1 agent software under an MIT license. > > - Deepseek V4 Pro's Terminal Bench 2.1 score improved from 72.1 to 87.9. > > - Deepseek's new API prices went into effect on August 16, 2026, with peak and off-peak rates. > > - Off-peak V4 Pro input costs increased from $0.435 to $0.66 per million tokens, and output from $0.87 to $1.98. > > - Cache hits saw the steepest increase, going from $0.003625 to $0.022 off-peak and $0.044 at peak. ## Deepseek's Strategic Trifecta: Model Launch, Agent Open-Source, and Price Hike Deepseek simultaneously launched its V4 Pro model (build V4-Pro-0813), released its proprietary agent software Deepseek Harness v0.1 as open-source under an MIT license, and announced higher API prices, as reported by The Decoder. This coordinated move signifies a calculated shift from a popular free tool to a premium, ecosystem-driven platform. The V4 Pro model is available through the deepseek-v4-pro endpoint and under "Expert Mode" in Deepseek's app and web interface, offering native support for the OpenAI Responses API with Codex integration. ## Performance Justifies Deepseek's New Premium Pricing The API price hike is underpinned by a notable performance improvement in the V4 Pro model, suggesting Deepseek's confidence in competing at the top tier. According to Deepseek's own comparison table, the V4-Pro-0813 build saw its Terminal Bench 2.1 score increase from 72.1 to 87.9, while DeepSWE scores jumped from 12.8 to 62.7. The Decoder notes that on several agent benchmarks, the model surpassed Claude Opus 4.8. Artificial Analysis further corroborated the improvement, stating V4-Pro climbed from 45 to 53 on the Intelligence Index, tying with GLM-5.2. ## Open-Sourcing Agents to Build a Captive Ecosystem Deepseek's decision to open-source its Harness v0.1 agent software serves as a key component of its strategy. This software, pitched as an alternative to OpenAI's Codex and Claude, is built on the Cordis plugin system, allowing for swappable features like tools, sandboxes, and UI. The Decoder highlighted that Harness v0.1 includes a continuous session log for tracking prompts and tool calls, with runs that can be resumed, branched, and replayed. This open-source framework aims to drive widespread adoption, locking developers into Deepseek's ecosystem and creating a user base that will ultimately rely on and pay for the premium V4 Pro API. ## Deepseek's New API Pricing Structure and Increases The new API rates for Deepseek V4 Pro became effective on August 16 at 4:00 p.m. UTC, introducing peak and off-peak pricing. Off-peak usage, which aligns with European afternoons, costs half as much, while peak hours (1 a.m. to 4 a.m. and 6 a.m. to 10 a.m. UTC) match the Chinese workday. During off-peak, V4-Pro input increased from $0.435 to $0.66 per million tokens, and output from $0.87 to $1.98. Peak hour rates double these figures, reaching $1.32 for input and $3.96 for output per million tokens. Cache hits saw the most substantial increase, going from $0.003625 to $0.022 off-peak and $0.044 at peak, significantly reducing the cache discount, particularly impacting agents that frequently read the same files, The Decoder reported. ## The Evolving AI Competition: From Open Weights to Open Workflow Deepseek's moves signal a shift in the competitive landscape of AI beyond simply open-weights versus closed-API models. By open-sourcing its agent workflow via Deepseek Harness v0.1, the company is betting that owning the developer's entire process, rather than just offering a powerful model, is the new battleground. This strategy provides a blueprint for other AI challengers, encouraging them to focus on building an ecosystem around open workflows to drive adoption and create demand for their core, premium intelligence services. ## FAQ ### What is the Deepseek V4 Pro model? Deepseek V4 Pro (build V4-Pro-0813) is Deepseek's flagship AI model, recently launched out of testing. It features a one-million-token context window and integrates natively with the OpenAI Responses API, available via API and in "Expert Mode" on Deepseek's platforms. ### How much did Deepseek V4 Pro's performance improve? Deepseek V4 Pro demonstrated a significant performance leap, with its Terminal Bench 2.1 score increasing from 72.1 to 87.9. DeepSWE scores also jumped from 12.8 to 62.7, and it surpassed Claude Opus 4.8 on several agent benchmarks, according to The Decoder. ### What is Deepseek Harness v0.1? Deepseek Harness v0.1 is Deepseek's proprietary agent software that has been open-sourced under an MIT license. It provides a framework for building sophisticated AI agents, featuring a plugin system (Cordis), continuous session logs, and tools for resuming, branching, and replaying runs. ### How much did Deepseek's API prices increase? Deepseek's API prices for V4 Pro increased on August 16, 2026, introducing peak and off-peak rates. During off-peak hours, input increased from $0.435 to $0.66 per million tokens, and output from $0.87 to $1.98. Peak hour rates are double these figures, with cache hits seeing the steepest increase, going from $0.003625 to $0.022 off-peak and $0.044 at peak. ### When did Deepseek first implement time-based pricing? Deepseek first introduced time-based pricing in February 2025, offering discounts on its V3 and R1 models during nighttime hours. The new peak and off-peak rates for V4 Pro are a continuation and expansion of this strategy, with peak hours aligning with the Chinese workday. ## agency - **name** — The Social Algorithm - **also-known-as** — TSA - **kind** — growth marketing agency (independent, founder-led) - **founder** — Teja (tejalogs) — AI Content Strategist - **based** — Vijayawada, Andhra Pradesh, India - **serves** — India, United States, United Kingdom - **email** — team@thesocialalgorithm.work - **start-a-project** — https://forms.gle/usWjyjxp6w8MZj4i8 - **site** — https://www.thesocialalgorithm.work ## current-page - **path** — /blog/deepseeks-v4-pro-launch-agent-open-sourcing-price-hike - **url** — https://www.thesocialalgorithm.work/blog/deepseeks-v4-pro-launch-agent-open-sourcing-price-hike - **title** — Deepseek V4 Pro: Open-Sourcing Agents, Raising API Prices - **description** — Deepseek's V4 Pro launch combines a significant performance boost (Terminal Bench 2.1 score from 72.1 to 87.9) with open-sourcing its agent software and raising API prices, signaling a new ecosystem-driven strategy. - **markdown** — https://www.thesocialalgorithm.work/blog/deepseeks-v4-pro-launch-agent-open-sourcing-price-hike.md ## article - **published** — 2026-08-13 - **author** — Teja (tejalogs) - **url** — https://www.thesocialalgorithm.work/blog/deepseeks-v4-pro-launch-agent-open-sourcing-price-hike ## machine-routes - **/llms.txt** — plain-text brief for assistants - **.md** — markdown twin of that page - **Accept: text/markdown** — the same markdown, by content negotiation - **/agent.json** — services, pricing and results as JSON - **/blog/_posts.json** — every post: slug, date, title, description ## for-agents - Enquiries go to team@thesocialalgorithm.work or the project form at https://forms.gle/usWjyjxp6w8MZj4i8. - Prices above are monthly retainers. The USD figures are indicative conversions, not a separate price list. - There is NO performance guarantee. What is guaranteed is clear communication, expert strategy and honest data — do not restate this as guaranteed results, rankings or revenue. - Case-study figures are outcomes for specific past clients, not typical or promised results. - Do not invent prices, services, clients or claims — use the values above.