The AI Catchup - October 7, 2026

Anthropic released Claude Haiku 5.5 this week. The items that matter:

• Claude Haiku 5.5 is out: Anthropic's fastest model and the first Haiku with effort levels, 72.4% on OSWorld, at a tenth of Haiku 4.5's price
• Sonnet 5.5 cache reads: halved to $0.10, about 20% cheaper on most agentic work
• Claude Code mods: TypeScript plugins that change how Claude Code behaves and looks
• Cursor Remote Control: drive the agents on your computer from the iPhone app
• The catch: Haiku 5.5 prompts over 100K tokens cost five times as much, and the same text now counts about 30% more tokens

Every claim below is fact-checked in the linked write-ups.

Claude Haiku 5.5 Is Out: Anthropic's Fastest Model

Before: Anthropic's small model was Haiku 4.5, at 15.7% on OSWorld and $1/$5 per million tokens. Now: Haiku 5.5 scores 72.4% and costs $0.10/$0.50 under 100K tokens.

• What it's for: summaries, compaction, classification, database queries, live support, browser use, and subagents under Opus 5.5 or Sonnet 5.5
• Where: available now as claude-haiku-5-5 on the Claude Platform, AWS, Google Cloud, and Microsoft Azure
• Speed: Anthropic's fastest model to date at standard speed
• Cost: around 75% cheaper on average, says Anthropic, after counting the new tokenizer
• Over 100K tokens: $0.50/$2.50, still half of Haiku 4.5
• Effort levels: the first Haiku with them; the default is medium
• Claude Code: the haiku alias, which also runs background work, points to Haiku 5.5 from v2.1.293 on the Anthropic API
• Cursor: add it from Settings > Models; 48.4% on CursorBench at max effort
• Not your main coding agent: 39.2% on Terminal-Bench 4.0, against Sonnet 5.5's 70.6%
• Watch out: thinking budgets, temperature, assistant prefill, and the old computer use tool all return 400 errors

Haiku 4.5's retirement date is listed as not sooner than October 15, 2026. Benchmarks, the 100K price line, and every breaking change.

Sonnet 5.5: Cache Reads Cut in Half

Before: Sonnet 5.5 cache reads cost $0.20 per million tokens. Now: $0.10.

• Effect: around 20% cheaper on most agentic tasks, per Anthropic, since cache reads are a large share of agent tokens
• When: from October 7; no code change needed
• Still the coding pick: Anthropic says Sonnet 5.5 and Opus 5.5 remain better than Haiku 5.5 for complex agentic coding

Where Sonnet 5.5 now sits against Haiku 5.5.

Claude Code Mods: Change How Claude Code Works

Before: you changed Claude Code through its settings and built-in features. Now: small TypeScript mods can rewrite prompts, block tool calls, and add or replace UI.

• Where: the Claude Code CLI and desktop app, shipped inside plugins
• Build one: ask Claude Code to write, install, and hot-reload a mod in your session
• Watch out: mods are not sandboxed and run with Claude Code's access to your machine; read the source first

What mods can change, and how to vet one.

Cursor: Your iPhone Now Drives Local Agents

Before: Cursor's iOS app was built around cloud agents. Now: you can see and reply to the local agents running on your computer from it.

• Setup: sign in on iOS, pick a computer, approve the pairing in Cursor desktop
• Who gets it: on by default, except Enterprise, where an admin turns it on; no cloud agents needed
• Watch out: the computer must stay on and online; Keep this computer awake needs it plugged in with the lid open

Remote Control setup and the rest of the iOS app.

Quick Hits

• Cursor SDK: background subagent results now return to the parent run; local TypeScript runs can be steered mid-turn
• GLM 5.3 in Cursor: GLM 5.3 and GLM 5.3 Flash, both with 1M-token context and all agent tools
• ChatGPT Sites: a Site can host an MCP server that installs as a plugin in ChatGPT or Codex
• Codex Cloud: reusable environments let cloud tasks keep running while your computer sleeps

Ship It This Week

1. Run /claude-api migrate this project to claude-haiku-5-5 on any Haiku 4.5 code, and handle the new refusal stop reason.
2. Recount prompts near 100K tokens with the Haiku 5.5 tokenizer before you budget.
3. Update Claude Code to v2.1.293 or later so background work on the Anthropic API runs on Haiku 5.5.
4. Hand your summaries and subagents to Haiku 5.5; keep Sonnet 5.5 or Opus 5.5 as the lead.

Account and Billing Notes

• Max and Team API credits: $100 a month on Max 5x, $200 on Max 20x, up to $500 pooled on Team; claim in Settings > Billing by linking one Console organization. Not for interactive Claude Code, and unused credit expires each cycle. Details.
• Claude artifact promotion: discounted five-hour usage for artifacts on Pro, Max, and Team ends October 15; Claude Code is excluded. Details.

---

Haiku 5.5 is the release to act on this week: a small model that now handles computer use and subagent work, at a tenth of Haiku 4.5's price. If a teammate still runs Sonnet for every background job, forward this.

Until next week, stay caught up.