Promptea Weekly
Week of 2026-09-20 to 2026-09-26
The big story
Anthropic ships Claude Opus 5.5, and the headline is the price
The new Opus lands at $4 and $20 per million input and output tokens, down from $5 and $25, with Anthropic claiming Fable 5.1-level results on most work. It is also the company's first model since Dario Amodei argued publicly for pacing the frontier.
This week in AI
Anthropic ships Claude Opus 5.5, and the headline is the price
The new Opus lands at $4 and $20 per million input and output tokens, down from $5 and $25, with Anthropic claiming Fable 5.1-level results on most work. It is also the company's first model since Dario Amodei argued publicly for pacing the frontier.
A 20% list-price cut at the frontier tier, plus a 60% cut on cache reads, changes the arithmetic for anyone running agentic coding or long-horizon workloads at volume — and pressures competitors' pricing at the same tier.
Read the full story →
Appeals court lets the Pentagon exclude Claude, and says guardrails can count as a supply-chain risk
A split D.C. Circuit panel ruled that Anthropic's refusal-by-training is 'manipulation' under a 2018 procurement law, whatever the company's motive.
It is the first appellate merits ruling in the Anthropic–Pentagon dispute, and it goes against Anthropic on every claim.
Read the full story →
Microsoft rebuilds Copilot around Home, Code and Autopilot, and puts the agents on metered billing
The new Copilot app merges Chat and Cowork, lets non-developers build apps that run inside the tenant, and adds an always-on agent. Most of it is still weeks away, and the agentic parts bill by usage.
Microsoft is formalizing a two-tier model for workplace AI: a flat per-user license for assistance and metered Copilot Credits for agents, which changes how organizations need to budget and govern Copilot.
Read the full story →
Claude flagged a CRISPR-shaped enzyme system. Nobody knows what it does.
Anthropic's new life sciences lab says ~950 Claude agents burned 21 hours and 210 million tokens to surface ART, a reverse transcriptase sitting next to a CRISPR-like repeat array. Its function is unestablished, the underlying enzyme was already known, and humans did every experiment.
A structural resemblance to CRISPR is a hypothesis, not a capability. Anthropic states outright that ART's function is unknown and the work to determine it is ongoing.
Read the full story →
OpenAI's new math advisory group can say anything. It can't slow anything down.
Nine mathematicians, hosted at the Institute for Advanced Study, can assess OpenAI's results and go public with objections. The charter explicitly excludes advising the company on how fast to move — the exact complaint twenty-five Fields Medalists made ten days ago.
The objection was about pace, and the remedy explicitly excludes pace. OpenAI's charter rules out advice on how fast its mathematics research moves, which is the exact grievance in the Fields Medalists' letter of 11 September.
Read the full story →
Google's Gemini avatars go GA, and the video bills at 6,192 tokens a second
Gemini 3.8 Live with Live Avatar is generally available in Gemini Enterprise. Google published the token conversion rates, which makes the cost of visual presence unusually easy to calculate — about $0.37 per minute of speech, roughly 21 times the audio it accompanies.
Google published per-second token conversion rates for avatar video, which makes this one of the few multimodal features you can cost out precisely before you build on it — 6,192 tokens per second of speech against a $1.00 per million rate.
Read the full story →