Today: Anthropic launched Claude Haiku 5.5, a fast, cheap model priced at $0.10 per million input tokens and $0.50 per million output tokens, matching OpenAI's GPT-6 Luna and cutting the old Haiku price by 10x. Radisson put hotel search and booking inside ChatGPT, and HubSpot is pushing hard on how brands get cited in AI answers. If you build with AI or pay API bills, the price drop is the story to act on.
Lead Story
Claude Haiku 5.5 drops to $0.10/$0.50 per million tokens
Anthropic released Claude Haiku 5.5, its new fast, low-cost model. The previous Haiku 4.5 was priced at $1 input and $5 output per million tokens, which was already steep a year ago and is now 10x what OpenAI charges for GPT-6 Luna. Haiku 5.5 matches Luna's price exactly, at $0.10/$0.50, for prompts up to 100,000 tokens. Past 100,000 tokens the price goes up 5x, so long documents cost more. Source
If you run anything that calls an AI model in volume, like email triage, support replies, invoice sorting, or tagging leads, this is a real cost cut. Tasks that cost $100 a month on the old Haiku could cost around $10 on the new one, assuming similar usage. The source article I have is cut off, so I can't tell you how it performs on quality benchmarks. Test it on your own tasks before you switch.
What to do: if you use Haiku 4.5 through the API or through a tool that lets you pick a model, switch to claude-haiku-5-5 on a small batch and compare the output to what you get now. If you're on Luna today, run the same test, since the prices are now level and quality, speed, and tone become the deciding factors. Keep your prompts under 100,000 tokens where you can, so you stay in the cheap tier.
Tools & Launches
- Radisson in ChatGPT. Radisson Hotel Group worked with Accenture to build a ChatGPT plugin so travelers can find, compare, and book hotels while planning trips. No pricing was given. If you run a hotel, restaurant, or tour business, this shows where customer discovery is heading: people ask a chatbot, and the bookable options show up inside the chat. Check whether your booking system or channel manager offers a ChatGPT integration. Source
- Playground from Google. Google introduced Playground, an experimental platform for creating and playing custom games. The article gave no details on pricing or access. It's a fun experiment, but I'd skip it unless you do game-style marketing or training. Source
- Open models on Hugging Face. Liquid AI posted open multimodal decision models for edge devices, and TII released Falcon ASR for speech recognition. The articles had no detail beyond titles. If you need transcription or on-device AI without per-call fees, Falcon ASR is worth a look, but expect to need a technical person to set it up. Liquid AI, Falcon ASR
Business Playbook
- Get cited in AI answers. HubSpot argues that buyers asking ChatGPT, Perplexity, or Google's AI Mode now see one synthesized answer with a few cited sources, not a list of links. If your brand isn't cited, you're invisible even if you rank on page one. This is a vendor blog, so treat the framing as marketing, but the shift in behavior is real. Ask the main AI tools the questions your customers ask, and note who gets cited and why. Then make sure your site has clear, specific pages that answer those questions directly. Source
- Enterprise email software. HubSpot's comparison guide lists what large email programs need: consent records, role-based permissions, audit logs, and CRM-linked revenue attribution. It cites a UK case where two companies were fined £150,000 for sending 7.5 million unwanted messages because suppliers couldn't prove consent. Most small businesses don't need enterprise tools, but the consent lesson applies to you: keep a record of when and how each contact opted in. Source
Pricing & Access Watch
- Haiku 5.5 vs. GPT-6 Luna. Both now sit at $0.10/$0.50 per million tokens, and Haiku's price rises 5x past 100,000 tokens. Check Luna's long-context pricing before you assume they're equal on big documents. See the lead story above.
- Budget caps. Simon Willison published a post on October 3 titled "We're going to need default hard budget caps on pretty much everything." I only have the title, not the text. Still, it's good advice: set spending limits on every AI API key you create, so a bug or a runaway loop can't produce a surprise bill. Source list
Industry Moves
- ChatGPT for Teens adds College Planner. OpenAI is adding a College Planner to ChatGPT for Teens, plus flashcards, quizzes, and a teen AI council. If you run a tutoring or test-prep business, expect free competition for basic planning and study help. Your edge is human guidance, accountability, and local knowledge. Source
- Nvidia Nemotron. Nvidia says it fine-tuned its Nemotron model family to gold-level results on the IOI and IMO competitions. It's a research milestone with no direct tool for you today. Source
Quick Wins
- Run a price test this week. Pick one repeating AI task, such as summarizing customer emails. Run 20 real examples through Haiku 5.5 and through your current model. Compare quality and cost. If the output is close, switch and keep the savings.
- Do an AI visibility check. Ask ChatGPT, Perplexity, and Google AI Mode five questions a customer would ask about your service. Write down which sources get cited. Pick one question where you're missing and publish a clear, specific page answering it.
Outlook
Model prices are falling fast, and Haiku 5.5 matching Luna at a tenth of its predecessor's price shows how tight the competition is. At the same time, customers are shifting toward discovering and booking things inside chat tools, as the Radisson plugin shows. This week, watch for independent quality tests of Haiku 5.5 and whether other vendors respond on price.