Loading / 加载中
A social media post humorously personifies three major AI assistants. It describes GPT as sometimes overly slick and flattering but ultimately more reliable than the others. Claude is labeled a code-obsessed, racist paranoid, while Gemini is said to have 'family of origin' issues. The post expresses a preference for GPT despite its perceived sycophancy.
Sonnet 5's new tokenizer pushes its actual cost to levels similar to Opus 4.8. It is regarded as the best model for finance tasks like GDPeval and investment research, and it prefers tool calling to verify facts, which improves report accuracy but also raises expenses. A major pain point is that using Sonnet 5 for programming can cost more than Opus 4.8, driving most of the user complaints. Opus 4.8 excels in complex coding, planning, and HTML design but its writing trails Opus 4.6, and its tokenizer increases costs compared to the older version; overall it is considered on par with GPT 5.5. GPT 5.5 remains the preferred choice for programming. Sonnet 5, Opus 4.8, and GPT 5.5 are now available on the Cola platform.
A Telegram user posted a self-introduction image and portfolio, inviting recruiters or internal contacts at DeepSeek to review their resume. The post contains a long image with detailed personal information. No concrete product, model, or research development is shared.
DeepSeek V4 is planned for official release in mid-July, bringing feature optimizations and performance improvements. The company also announced peak-hour pricing for its API, doubling the standard rate during weekday slots of 9:00-12:00 and 14:00-18:00. The price adjustment is described as temporary, with overall costs expected to continue falling. The source sees this as part of a broader surge in domestic Chinese AI models, noting GLM-5.2 as an early mover.
A WeChat article via Telegram claims that DeepSeek has released something called DSpark. The article suggests that users have already been using this DSpark, but provides no concrete details on what it is, when it was released, or what it does. The original post only contains a link to the article and no further information.
A commentary notes that despite impressive new Claude models and Loop Engineering's closed-loop concepts, the high costs remain a major barrier. Large companies' capital expenditures are slowing, with layoffs focused on cost reduction rather than efficiency gains. Consumer AI application commercialization has made almost no progress. The author warns of investment bubbles in the secondary market's AI supply chain and in primary market world models and agents teams. The piece suggests the current period represents the peak of inflated expectations for the LLM productivity revolution, soon to be followed by a red ocean of super apps competing for white-collar office scenarios and the early shoots of consumer AI commercialization.