DeepSeek simply launched its strongest AI mannequin but, aiming instantly at rivals with higher efficiency and far decrease pricing.
The Chinese synthetic intelligence startup on Thursday launched the official model of its flagship mannequin, DeepSeek-V4-Pro-0813, throughout its net interface, cell app, and API. Designed to excel at autonomous AI agent duties and software program engineering, the discharge delivers capabilities aggressive with main fashions like Moonshot AI’s Kimi K3, however at a fraction of the fee.
The general-availability launch marks a considerable improve over the preview model launched in April. According to DeepSeek, the brand new V4 Pro achieved a rating of 62.7 on the DeepSWE benchmark for real-world software program engineering, up sharply from 12.eight in preview testing. It additionally recorded an 87.9 on Terminal Bench 2.1.
Independent evaluations spotlight DeepSeek’s technical progress alongside its low pricing. Independent benchmarking agency Artificial Analysis awarded the reasoning model of V4 Pro a rating of 53 on its Intelligence Index. While that trails Moonshot’s Kimi K3 rating of 60 and Anthropic’s Claude Opus 5 rating of 63, DeepSeek’s pricing considerably undercuts its rivals.
On third-party platforms like OpenRouter, V4 Pro is listed at $0.435 per million enter tokens and $0.87 per million output tokens. In distinction, Moonshot’s Kimi K3 prices $14 to $15 per million output tokens, whereas U.S. opponents similar to Anthropic’s Claude Fable 5 run as excessive as $50 per million.
DeepSeek additionally launched DeepSeek Harness (dsh) v0.1 beneath the open-source MIT license, offering an alternative choice to Anthropic’s Claude Code. Built on a modular structure during which “everything is a plugin,” the device lets builders orchestrate autonomous coding brokers throughout native file techniques and growth environments.
Alongside these launches, DeepSeek introduced an overhaul to its API billing. The firm will exchange flat charges with peak and off-peak pricing, elevating API prices between 50% and 1,100% relying on mannequin kind and utilization hours. During peak hours, V4 Pro output will climb to $3.96 per million tokens, a pointy leap from present charges, however nonetheless under Western alternate options.
What this implies for builders and companies
DeepSeek’s technique is changing into extra nuanced. The firm is elevating API costs, however additionally it is attempting to persuade builders that V4-Pro delivers sufficient functionality to switch far more costly fashions in manufacturing techniques.
That might matter for startups and enterprise groups working large-scale AI workloads, the place inference prices typically decide which mannequin will get deployed. A mannequin that’s barely weaker on some benchmarks however dramatically cheaper to function can nonetheless be the higher enterprise resolution.
The tradeoff is that DeepSeek is now not competing solely on ultra-low pricing. By shifting V4-Pro right into a premium tier whereas conserving it nicely under many U.S. opponents, the corporate is testing whether or not it might probably construct a sustainable enterprise somewhat than merely win a value struggle.
Related studying: For extra on DeepSeek’s broader ambitions, see how the corporate is pushing into the AI chip race as China works to strengthen its place within the world semiconductor market.
