AI's on the council now

gm legends, happy Friday.

Claude is going after long messy work across code and documents, Perplexity’s Model Council lets you stack top models against the same question, and OpenAI Frontier is trying to be where big companies park their agents once they are allowed to touch real workflows.

Claude Opus 4.6 is Anthropic’s new flagship model for long, messy tasks across code, documents, spreadsheets and analysis. It adds a 1M token context window, stronger planning for agent workflows, and better performance in finance, legal, security and general reasoning, while keeping Opus pricing in the same bracket. It is available in the Claude apps and via API for teams that want one model for both coding and knowledge work.

🔥 Our Take: The big labs are clearly aiming at the work that justifies salaries and software budgets, not toy prompts. This release goes straight at that layer: huge repos, financial research, legal tasks and security reviews that used to sit in specialist tools. If it holds up outside neat benchmark runs, a lot of companies will quietly start pushing more of their week through it.

Samarth started a thread inviting founders to share their YC rejections and what they learned from them. The prompt is simple: what reason did you get, and what would you tell future applicants to do differently with their idea, profile, or application so they have a better shot.

If you have a “we bombed, here is why” story and a few hard earned tips for the next batch, this is where to drop it.

Model Council in Perplexity runs your question through three frontier models at once, compares what they say, and then gives you a single combined reply. You can see where they agree or disagree, check the evidence, and swap models in and out when you want a closer look. It is available on web for Perplexity Max users.

🔥 Our Take: People already copy paste the same question into a few tabs to see if everyone says the same thing. This just does that in one place with less hassle. For anything important, a quick cross check like this feels a lot more honest than pretending one model is always right.

OpenAI Frontier is a platform for companies to run AI agents on real workflows instead of scattered experiments. It plugs into things like CRMs, data warehouses and internal tools so agents work with proper context, permissions and audit trails. It also supports non-OpenAI agents so everything shows up in one view.

🔥 Our Take: Big companies already have random bots wired into different tools with nobody really owning the whole mess. This tries to turn that into one setup where you can see what agents exist, what they touch and whether they are actually useful. It will earn its keep if it cuts down on custom glue and mystery scripts, not because the diagrams look nice.