DeepSeek V4 API shifts to time-based pricing, doubling costs during peak UTC hours

·1 views

DeepSeek switched its entire API to peak and off-peak billing on August 16, 2026, replacing a flat-rate pricing model. Under the new structure, output from the flagship V4-Pro model costs $3.96 per million tokens during peak hours and $1.98 off-peak, compared to a previous flat rate of $0.87. Peak hours are defined as 01:00–04:00 and 06:00–10:00 UTC — windows that fall squarely during standard working mornings in the UK and Central Europe, while largely missing US business hours. The most dramatic increase is in cached input pricing, which rose roughly 1,100% at peak for V4-Pro, disproportionately affecting applications that repeatedly send large system prompts. DeepSeek says the restructuring is intended to manage infrastructure capacity and encourage users to schedule workloads around actual demand.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Gooseworks, YC W23 Startup, Seeks Founding Engineer

Gooseworks, a startup that participated in Y Combinator's Winter 2023 batch, is actively recruiting for a Founding Builder or Engineer role. The position was listed on Y Combinator's job board, signaling an early-stage hiring push. As a founding engineer, the hire would likely play a pivotal role in shaping the company's core product and technical direction. The listing had no public comments or engagement points at the time of posting.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

How One Developer Replaced ChatGPT, Claude, and Perplexity With a Single AI Orchestrator

A developer cancelled paid subscriptions to ChatGPT Pro, Claude Pro, and Perplexity Pro, replacing them with a single tool called Littlebird that acts as an orchestration layer across AI models. Rather than logging into multiple platforms separately, the developer uses Littlebird as a unified front end that routes tasks to appropriate underlying models based on context. The setup is anchored by a Mac Studio, which serves as a local hub storing files, notes, and project data that Littlebird draws on when sending context to external AI tools. Results returned by those external models are written back to the Mac Studio, allowing Littlebird to maintain continuity across conversations without relying on any single vendor's storage. The approach is framed as a way to reduce subscription costs and tool sprawl while keeping data and workflows under personal control.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How ABM Landing Pages Outperform Generic Homepages in Maritime Cybersecurity

Account-based marketing (ABM) landing pages require a fundamentally different approach than generic homepages, particularly in niche industrial sectors like maritime cybersecurity. Using Blackhole Networks as a case study, the analysis highlights how generic homepages fail ABM traffic by lacking focused lead capture, relevant social proof, and sector-specific copy. Rather than directing targeted visitors to a broad homepage, effective ABM pages function more like product pages with a single clear call to action tailored to roles such as port IT leads or fleet security officers. Social proof on these pages should go beyond logo strips, referencing outcomes specific to maritime environments like OT networks, regulatory compliance, and incident response. The piece argues that copy must move away from generic terms like 'digital transformation' and instead reflect the precise language and risk concerns of the maritime sector.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Blogger audits 200 posts, finds three hidden bugs in headings, CSS, and mobile layout

A developer running a long-standing technical blog paused new publishing to conduct a full audit of roughly 200 existing posts, uncovering three structural issues invisible on their own screen. Multiple posts contained conflicting H1 headings caused by the blog theme and post body editor rendering titles independently, resulting in eight duplicate top-level headings across five posts. A screen-reader label on the site's search box was marked with a CSS class intended to visually hide it, but the corresponding style rule was missing entirely, causing layout breakage on narrow mobile viewports. The correct fix used a clip-based CSS pattern to keep the label accessible to assistive technology while removing it from visual display — rather than simply deleting the element. Each fix was deployed as a separate versioned theme update with a recorded checksum, allowing for clean rollback if needed.

0 comments Read more at DEV Community