Claude's cheaper Haiku and a new ChatGPT interface
Anthropic says its new Haiku 5.5 model costs around 75% less to run than Haiku 4.5, making it worth testing for repetitive work. GPT-6 is rolling out in ChatGPT with interactive charts and forms, while eligible startups can get a free year of Claude Team and $1,000 in API credits. We also look at an offline meeting note-taker and what to check before connecting AI to company data.
The Big Story
Anthropic News
Introducing Claude Haiku 5.5
Anthropic released Claude Haiku 5.5 for high-volume tasks such as summarization and customer support, saying it costs around 75% less to run on average than Haiku 4.5. For prompts up to 100,000 tokens, API pricing is $0.10 per million input tokens and $0.50 per million output tokens (higher rates apply above that threshold). It’s available on Claude’s platform and AWS, Google Cloud, and Azure; Anthropic also halved Sonnet 5.5 cache-read prices and plans monthly API credits of $100 to $200 for Max subscribers and up to $500 pooled for Team users.
If your business pays for repetitive AI tasks, test Haiku 5.5 against your current model for quality and total cost, especially if prompts might exceed 100,000 tokens.
For Everyone
OpenAI News
GPT-6 and Intelligent UI for everyone
OpenAI is rolling GPT-6 out to ChatGPT users with Intelligent UI, which can build interactive charts, forms, diagrams, and small tools directly in a conversation. OpenAI says GPT-6 Instant starts answering web-search questions 44% sooner on average than GPT-5.6 Instant in its internal tests. The rollout begins October 7 for Plus, Pro, Business, and Enterprise, then October 8 for Free and Go; Enterprise access depends on admin settings, and Work and Codex models aren't changing.
Test the new Chat interface on a real planning or analysis task, but don't assume this release upgrades existing Work or Codex workflows.
TechCrunch AI
Google releases a new local-first Granola competitor
Google released AI Edge Foresight, a Mac meeting note-taker that can transcribe conversations and generate notes entirely offline. It lets users keep their own notes alongside AI-generated ones, ask questions about transcripts, and add documents to a searchable knowledge base. Google says it's optimized for Apple Silicon; TechCrunch notes the app may be more a demonstration of Google's on-device models than a mainstream product.
If meeting privacy or unreliable internet makes cloud note-takers a poor fit, test Foresight on an Apple Silicon Mac before committing to it for team workflows.
For Builders
GitHub Trending
cathrynlavery/diagram-design: Editorial diagram design for Claude Code, Codex, GitHub Copilot, Factory Droid, and Pi. 42 diagram types. Self-contained HTML + SVG. No shadows. No Mermaid slop.
The open-source Diagram Design skill generates 42 kinds of diagrams, from flowcharts and timelines to architecture and data-model views, as standalone HTML and SVG. It works with Claude Code, Codex, GitHub Copilot, Factory Droid, Pi, and other compatible agent tools, and can redraw Mermaid, draw.io, or Excalidraw diagrams. Each type includes light, dark, and editorial variants that open in a browser without a build step.
If you regularly make process or architecture diagrams, inspect the live gallery before trying the skill, and review its privacy documentation before allowing it to read your website or source diagrams.
For Business
TechCrunch AI
Anthropic is giving startups a free year of Claude Team and $1,000 in credits
Anthropic expanded Claude for Startups with a free year of Claude Team for up to five premium seats and $1,000 in API credits. Eligible companies must have been founded within the past five years or funded within the past two; they can also access Claude Marketplace and book virtual office hours with Anthropic's Applied AI team. Applications are available through the program page.
If the company meets the eligibility rule and already plans to test Claude, apply and check what the seats and API usage will cost after the offer ends.
OpenAI News
Atlassian and OpenAI expand their partnership
Atlassian and OpenAI expanded their partnership to put GPT-6-family models into Atlassian’s platform and its Rovo assistant, which can use context from Jira, Confluence, and other work data. Atlassian also offers plugins that bring Jira items and Confluence content into ChatGPT and Codex, subject to permissions. Deeper Jira features for assigning work to AI agents and tracking their results are still being explored, not announced as available.
If your team already uses Jira or Confluence, check which Rovo and ChatGPT integrations your plan supports and review access permissions before connecting company information.
Worth Watching
TechCrunch AI
Mistral’s new 1T model aims to leapfrog closed and open rivals
Mistral released Mistral Large 4, a 1-trillion-parameter multimodal model currently accessible only through a guarded public endpoint. It plans to publish the model weights after ~three weeks of safety testing, which would let organizations inspect and run it themselves. Mistral points to cybersecurity, finance, and chip design as target applications, but independent benchmark results aren't available yet.
Wait for published weights, independent tests, and deployment costs before comparing it with models already in use.
TechCrunch AI
Nous Research confirms it hit $1.5B valuation, launches AI agents for business users
Nous Research raised $90 million at a $1.5 billion valuation and announced Hermes for Businesses, a way for companies to deploy customized AI agents for multistep work. The company says the business offering keeps data private and secure, but the article provides no pricing or independent assessment of those protections. Its open-source Hermes Agent is already popular with developers, according to Nous's usage estimates.
Businesses considering Hermes should ask for pricing, deployment details, and evidence of data controls before putting customer information into an agent workflow.