Product & Strategy
The Product Desk
Claude Haiku 5.5 costs a tenth as much per token and more per test than Haiku 4.5.
The price sheet sells a per-token discount, and about 3.5x more reasoning steps eat it before the invoice arrives. The new tokenizer also counts the same text as 25–30% more tokens, and past 100K tokens the rate jumps 5x. Whether a migration saves money or raises it comes down to one measurement: which side of that cliff your production prompts actually sit on.
In Play
The Cheap Tier Ties on Price, Not on Cost
Anthropic priced Claude Haiku 5.5 at $0.10/$0.50 per million tokens, the same as OpenAI's GPT-6 Luna. Vals still found it costs more per test than Haiku 4.5 on every shared benchmark, per AINews. Three things drive that: about 3.5x more reasoning steps, a tokenizer that counts roughly 25–30% more tokens, and a 5x rate above 100K tokens. How long your prompts run decides whether a migration saves money.
Ask ClarityRouters Pick the Model, Clones Copy the UI
Microsoft and SpaceX made the same move within 48 hours, The Information reports. Microsoft said Copilot will send some tasks to its local MAI-Code-1.1 model. SpaceX said Grok Bot will send requests to Claude Opus 5.5, Midjourney and Suno. Google's Gemini agent also picks a model for each job. ChatGPT's new Intelligent UI was reportedly reverse-engineered in under 24 hours. Neither the model nor the rendering layer protects your product now.
Ask ClarityAgents Arrive as Your Customers
Meta and Sierra launched the Personal Agent Protocol, an open standard for how personal AI agents sign in to and deal with businesses. Shopify, Stripe, Walmart, Genesys, Instinct and Rocket backed it. Cloudflare says agents drove 48% of usage of its Wrangler command-line tool last week, and it now lets agents pay per request. Your funnel, pricing and API are serving a non-human customer that your dashboards probably don't separate out.
Ask ClarityAI Value Claims Face an Audit
The FT reports OpenAI's annualized revenue was near $50B at end-September, not the widely reported $70B. The gap came from multiplying one month's revenue by 12. Separately, a survey by The Information found a third of tech-forward subscribers say AI returns are a multiple of spend. Another 47% are unconvinced, and Copilot usage is falling even as seat counts rise. Expect renewals to turn on value your product can prove from inside its own usage data.
Ask Clarity
Deep Dives
- ●
Haiku 5.5's Rate Card Hides Three Multipliers, and Your Prompt Lengths Decide the Bill
The cheapest model tier now ties on list price. Whether you save depends on reasoning effort, tokenizer inflation and how many prompts cross one cliff.
Three multipliers sit between the list price and the bill Anthropic discloses each of these in its own fine print. The finance partner usually sees them for the first time on the invoice. Tokenizer inflation. Haiku 5.5 counts the same…
3 action items
- ●
Routers Now Pick the Model and Clones Copy the UI. Here Is What You Still Own.
Two layers that looked defensible a month ago lost that status in one week. That leaves governance, callability and domain judgment as the ground you can hold.
Look at what each router hides from the user Router Who picks the model Named back ends Governance Where results land Grok Bot (SpaceX) Grok Bot; no model picker Claude Opus 5.5, Midjourney, Suno Team allowlist honored by default, but…
3 action items
- ●
The Next Customer in Your Checkout Is an Agent You Can't See Yet
Protocols, payment rails and usage data all point the same way, but no one has a loss baseline. Whoever defines safe agent purchasing first sets the category's terms.
The protocol coalition maps onto your funnel Meta brings the consumer side of the Personal Agent Protocol. Sierra and Genesys bring customer service, Shopify and Walmart bring storefronts, and Stripe brings payments. That covers discovery, checkout and support in one…
3 action items
The edition continues
Take the signal into the room.
Sign up or log in to read all 3 deep dives in full, plus the final take.
Read the full editionContinue with LinkedIn