Claude Haiku 5.5 ships: 1M context at $0.10/M, now the default Haiku
Anthropic's cheapest model yet is live on four clouds, with a migration catch. Plus free API credits for Max and Team, SynthID opens to all, OpenAI trims tiers.
Copy markdown
A 1M-context Haiku at $0.10 per million in
Claude Haiku 5.5 (`claude-haiku-5-5`) is Anthropic's cheapest, fastest model yet: a 1M-token context window, 128K max output, and roughly 75% lower run cost than Haiku 4.5, at $0.10/$0.50 per million tokens under 100K (then $0.50/$2.50 above that). It's live on the Claude Platform, AWS, Google Cloud and Azure, and is already the default Haiku in Claude Code 2.1.293 — so high-volume extraction, classification and cheap agent subtasks just got far cheaper to run.
It's the default now — and it may break your code
The upgrade isn't frictionless: adaptive thinking is on by default, so responses can open with thinking blocks; the old manual `budget_tokens` parameter now returns a 400 error; and the retokenizer counts slightly more tokens per prompt. Because the `haiku` alias resolves to 5.5, anything pinned to it inherits all of this silently — re-test any Haiku-dependent prompt before it ships.
Your Max or Team plan now funds your API bill
Anthropic is giving subscribers monthly Claude API credits to run their own apps and agents on the platform: $100/month on Max 5x, $200/month on Max 20x, and up to $500/month pooled across a Team. Link a Claude Console organization under Billing to claim it — enough to stand up a side agent without ever topping up a card.
Check if media is AI — now across OpenAI and Nvidia
Google's SynthID Detector is now public worldwide (English) at synthid.com, flagging SynthID watermarks in images, video and audio generated by Google (Nano Banana 2, Veo, Lyria), OpenAI, Nvidia and Kakao, with Apple expected next. It's not a general-purpose detector — it won't catch Claude output or locally-run open models, and a clean result proves nothing — but for anyone shipping generative features into the EU it's a free first pass at the AI Act's Article 50 labeling rule.
OpenAI's five usage tiers are now three
OpenAI collapsed its API usage tiers from five into three — Build, Launch and Grow — with organizations auto-promoted as cumulative credit purchases cross each threshold. Your rate limits may have quietly moved, so check your org's new tier before scheduling a big run.