PUBLIC INTELLIGENCE
Selected AI market intelligence
Recent moves by company. Sign up for the full timeline, company profiles and trend analysis.
DeepSeek Platform · major
DeepSeek Implements Weekend Off-Peak Pricing and Clarifies Deduction Rules
DeepSeek adjusts its billing rules, implementing all-day off-peak pricing on weekends (Beijing time) starting August 23, 2026, to encourage large-scale batch inference and smooth out compute demand. It also clarifies deduction rules, prioritizing the use of gifted balances over recharged funds to alleviate user concerns about locked capital.
DeepSeek Platform · milestone
DeepSeek API Introduces Peak/Off-Peak Pricing with Significant Adjustments
DeepSeek announced a shift to peak/off-peak pricing for its API services, with off-peak rates at half the peak rates. Peak hours are UTC 01:00-04:00 and 06:00-10:00, effective August 16, 2026. For deepseek-v4-flash, off-peak input is $0.22/M and output $0.66/M; peak input is $0.44/M and output $1.32/M, representing a significant price increase during peak hours.
DeepSeek Platform · major
DeepSeek V4 Series Model Specifications and API Compatibility Details
DeepSeek's official pricing page details the specifications and features of the V4 series models. The context window is uniformly expanded to 1M tokens with a maximum output of 384K. New Beta features include FIM (Fill-in-the-Middle) and Chat Prefix Completion, enhancing competitiveness in code assistant scenarios. It also supports Anthropic API format and Tool Calls, significantly reducing migration costs from the Claude ecosystem.
DeepSeek Platform · major
DeepSeek V4 Series Model Upgrades, Responses API Support, and Concurrency Limit Adjustments
DeepSeek has rolled out significant updates to its V4 series, advancing DeepSeek-V4-Pro to version 0813 and natively supporting the Responses API to streamline complex Agent development. Concurrent request limits remain fixed at 2,500 for V4-Flash and 500 for V4-Pro, while legacy footnotes were removed to clarify usage. These changes elevate inference performance and expand the Agent development ecosystem.
DeepSeek Platform · major
DeepSeek Releases Experimental Multimodal Vision Model V4-Flash-Vision-Exp
DeepSeek launched the experimental multimodal vision understanding model DeepSeek-V4-Flash-Vision-Exp, now available via its API platform. While matching DeepSeek-V4-Flash in pure text capabilities, it shows significant improvements in vision-based agent benchmarks, approaching Opus-4.8. This marks the first introduction of multimodal vision capabilities in the V4 series, released experimentally to gather rapid feedback. Pricing, rate limits, and official release dates are yet to be disclosed.
The above is a public selection. Sign up to access full timelines, per-source evidence and trend analysis.
Sign up for more intelligence →