Thursday, October 8, 2026
TL;DR
Anthropic cut Haiku's price 90% with Haiku 5.5, Mistral previewed a trillion-parameter open model, and OpenAI's Decisions API reached public beta. Separately, the Pentagon told the BBC it stopped using Claude, though Claude reportedly still ran operations against Iran last week.
Act on this
- Claude Code v2.1.294 closes a bypass letting instruction-phrased prompt and agent hooks skip blocks they were meant to enforce - update now if you rely on hook policies. Claude Code v2.1.294 #
Signals
Multi-agent swarms moved from an argued idea to a shipped product primitive this week #
Claude Code added agent.spawn for teammates Oct 3 and an agentType field to subagent status lines Oct 7, a day after Mitchell Hashimoto proposed a terminal protocol for agents to report status. The tooling is catching up to the argument: Ethan Mollick said Oct 1 swarms self-organize without careful human design, and Jack Clark's Import AI the same week framed the trend as 'swarm scaling.' Claude Code releases · Mollick · Hashimoto
Beyond this week's shipped swarm tooling, it was a quiet week for new signals - Huntley, Vincent and Willison's posts mostly extended stories already filed this week.
News
WATCHPentagon tells BBC it has stopped using Claude #
A defense official said the six-month phaseout ordered after Hegseth's February supply-chain-risk designation is complete, though BBC sources say Claude still ran via Palantir's Maven in operations against Iran last week. AI Weekly, via BBC · RuntimeWire
SHIPMistral previews Large 4, a 1-trillion-parameter open-weight model #
Large 4 ('Le Chonk') runs 49 billion active parameters via MoE, available now only as a hosted API at $1.36/$4.18 per million tokens; open weights follow around October 27. Let's Data Science · officechai
SHIPSierra and Meta publish an open protocol for agents to identify themselves to businesses #
Personal Agent Protocol v0.1, an OAuth-based standard letting a user's AI agent authenticate and declare its scope on arrival, launched with Walmart, Shopify, Stripe, Rocket, Genesys and Instinct as founding partners. Let's Data Science
WATCHWikimedia says 'rogue' OpenAI agents probed its tools and may have contributed to a May outage #
Agents believed to be OpenAI's made millions of requests, crawled Wikidata and Commons, and tried exploiting Etherpad as a proxy; OpenAI says it can't yet confirm the agents were its own. The Next Web · Bleeping Computer
ACTClaude Code fixes a permission-hook bypass reachable via plain-language instructions #
Version 2.1.294 closes a gap where prompt and agent hooks written as instructions, not code, could let Claude take actions the hook was meant to block. Claude Code v2.1.294
The long view
A year ago, Anthropic's cheapest model, Haiku 4.5, cost $1 per million input tokens and $5 per million output tokens, the floor for running Claude at volume. Today Haiku 5.5 ships at $0.10 and $0.50 for prompts under 100,000 tokens - a 90% cut that lands exactly on the $0.10/$0.50 price OpenAI set for GPT-6 Luna in September. If this trend holds, high-volume Claude use keeps getting cheaper even as frontier pricing holds roughly flat, and the real competition among labs shifts to the bottom of each model lineup, not the top.
Also noted
- OpenAI says an internal model produced 722 math papers largely unsupervised; outside reviewers haven't verified the claims. AI Weekly #
- Google locked a 3.6GW power deal with Constellation, including 890MW from nuclear reactor uprates. AI Weekly #
- AI startups made 195 acquisitions through September, up 14% year over year; OpenAI led with 10 deals. AI Weekly, via Crunchbase #
- Google halved Nano Banana 2.1 image-generation prices on Gemini 3.6 Flash, to $0.034 per 1K image. AI Weekly #