DeepSeek Launches V4-Pro AI Model with Agent Capabilities and Price Hike
DeepSeek launched its V4-Pro AI model on Thursday, now accessible via app, web, and API. This model, designated DeepSeek-V4-Pro-0813, emphasizes agent capabilities and has been in preview since April. It achieved notable benchmark scores, including 87.9 on Terminal Bench 2.1. The context window extends to 1 million tokens, allowing output of 384,000 tokens. A price increase takes effect on August 16, raising output token costs to $3.96 per million at peak times. Competitors in the space include Anthropic, Moonshot AI, Alibaba, and ByteDance, while DeepSeek recently secured $7.4 billion in funding.
Key facts
- DeepSeek officially released V4-Pro on Thursday, available on app, web, and API.
- Model was in preview since April.
- Designated DeepSeek-V4-Pro-0813, focuses on agent capabilities.
- Benchmark scores: 87.9 on Terminal Bench 2.1, 62.7 on DeepSWE, 61.5 on NL2Repo.
- Context window up to 1 million tokens, outputs up to 384,000 tokens.
- V4-Pro API updated to work with OpenAI Responses API format and includes Codex integration.
- Thinking effort levels expanded to low, high, and max for V4-Pro and V4-Flash.
- Price increase effective 16:00 UTC on August 16, with peak and off-peak billing.
- V4-Pro output tokens rise to $3.96 per million at peak from $0.87.
- Anthropic's Fable 5 charges $50 per million output tokens.
- DeepSeek had offered 75% promotional discount on V4-Pro through May 5.
- V4-Flash performed above expectations in independent tests.
- DeepSeek's R1 model went viral in early 2025.
- Competitors include Moonshot AI, Alibaba, and ByteDance.
- DeepSeek secured $7.4 billion in first outside capital round in June.
- Reuters reported discussions for additional round valuing at $74 billion.
Entities
Institutions
- DeepSeek
- OpenAI
- Anthropic
- Moonshot AI
- Alibaba
- ByteDance
- Reuters
- Global Times
Sources
- Quartz —