@scaling01 All I can say about this is that Anthropic should probably look into who has access to what and revoke certain accesses. Likely not Mythos tho
@pbakaus@OpenAI The fact that the agent skills spec did not think about subagents at all is one of the biggest misses in it. I have seen so many skills that have some agent definitions in the reference files the orchestrator is instructed to read and give as instructions to subagents
@Miles_Brundage They are absolutely fleecing enterprise customers, though, who pay normal API prices for their Claude code subs. Every model release it gets like $200-$500 more expensive per dev/month
@sporadica If the puts would have a 3 month expiration date, he would be down 97% on his puts since the Q1 reported numbers. If they expire in 1-2 years he would be down 50-70%. I think he just got burned betting that the large players where overvalued compared to the underlying infra
@Presidentlin Let's see how the news cycle plays out for this when the US boys wake up. Perfect time to release this after openai just doubled prices of their main model line
API is Available Today!
πΉ Keep base_url, just update model to deepseek-v4-pro or deepseek-v4-flash.
πΉ Supports OpenAI ChatCompletions & Anthropic APIs.
πΉ Both models support 1M context & dual modes (Thinking / Non-Thinking): https://t.co/MUPiwkDI8T
β οΈ Note: deepseek-chat & deepseek-reasoner will be fully retired and inaccessible after Jul 24th, 2026, 15:59 (UTC Time). (Currently routing to deepseek-v4-flash non-thinking/thinking).
6/n
@VictorTaelin For everyone who works with PowerPoint professionally, this has been the biggest improvement I have ever seen. This model might actually make strategy consultants feel like what coders have been feeling for the last year.
@Shpigford It's a good model. Also the first time where the loop "code UI -> take a screenshot -> fix styling bugs ->..." works really well. Haven't had text clipping or overflow since (also works amazingly with PowerPoint)
You should install all since they reference each other. Run teach-impeccable first in a repo to get it set up, after that run audit and critique, issues raised there will tell you what skills to activate to fix the issues. When building a new feature, just use frontend-design, it's by far the most powerful and comprehensive skill in the list
Github Copilot CLI is the most underrated agentic coding tool out there, especially if you like using subagents. Being able to launch a fleet of subagents with whatever frontier models you like + this not costing extra credits has completely changed my workflow