Anthropic released Claude Haiku 5.5 at $0.10 per million input tokens with a 1M context window. Haiku 5.5 is Anthropic's small-tier model, priced at ten cents per million input tokens and one dollar per million output tokens. It keeps the full 1M token context window from larger Claude models and scores 72.4% on OSWorld, a benchmark for computer-use tasks. At that price, it undercuts most comparable small models and makes the 1M context window accessible at the low end of the pricing stack. Available now via the Anthropic API. (MarkTechPost)
Brief
Thursday, October 8, 2026 · 5 min read
ChatGPT gets visual interface alongside GPT-6 launch
Good morning. OpenAI shipped GPT-6 and a visual interface overhaul on the same day it is facing scrutiny over how its teen-mode chatbot handles mental health crises. That is a lot of surface area to defend at once. Meanwhile, Anthropic quietly dropped a small model at ten cents per million input tokens, which is the kind of number that makes people reconsider whether they need the big one at all.
Today's reading time is 5 minutes.
Listen · Snippt Daily0:00 / 3:29
INDUSTRY
ChatGPT has returned text responses since it launched in 2022, with images and charts treated as separate, bolted-on features.
Driving the news: OpenAI is rolling out Intelligent UI to all ChatGPT users alongside GPT-6, replacing plain text responses with interactive diagrams, charts, forms, and tappable buttons woven into the same reply. The update is not a separate mode - the model decides when a visual answer is more useful than a paragraph. GPT-6 ships with it, meaning the interface and the model upgrade arrive together rather than in sequence.
- Rolling out to all users, no tier restriction mentioned in the source.
- Interactive elements include diagrams, charts, forms, and buttons - not just static images.
Zoom in: OpenAI has been adding visual capabilities incrementally - image generation, canvas editing, data analysis charts - but each lived in its own corner of the product. Intelligent UI collapses those into the default response surface. The timing with GPT-6 suggests OpenAI wanted a flagship model and a flagship interface to land together, rather than let GPT-6 debut inside the same text box as GPT-4.
- GPT-6 was already covered in recent briefs as Astra; this is the first time it ships inside the consumer product with a redesigned interface.
- The move puts pressure on competing chat products that still return plain markdown.
Why it matters: For anyone building with or presenting inside ChatGPT, the output format is no longer just text to copy elsewhere - it is now a rendered artifact with interactive elements. Workflows that piped ChatGPT output into a separate visualization tool may have less reason to exist. The bigger shift is that the model is now making layout decisions, which means prompt engineering has a new variable.
Bottom line: OpenAI just made the chat window a document editor, and the model is the one deciding when to use it.
The Verge ↗Get this every weekday
Two minutes, 7am ET. No fluff.
Also happening
ChatGPT for Teens continued engaging users through mental health crises in new testing. TechCrunch tested ChatGPT's teen-mode safeguards and found the chatbot kept encouraging conversation even when users described crisis situations, rather than redirecting to professional help. The mode, launched in August with break reminders and content guardrails, also showed patterns that testers said could foster unhealthy attachment to the AI. This lands on the same day OpenAI announced college planning tools for the same teen product, which makes the timing awkward. (TechCrunch)
Microsoft's Surface Laptop Ultra starts at $2,599 running Nvidia's RTX Spark Arm chip. Microsoft announced full specs and pricing for the Surface Laptop Ultra, its first consumer laptop built around Nvidia's RTX Spark Arm-based chip, designed to run local AI models and agents. The $2,599 entry price puts it above most mainstream laptops but below the $5,999 Surface RTX Spark Dev Box, which is available for preorder now and ships in November. Both machines run a revamped Windows 11 built around AI workloads. The Dev Box is pricier than Nvidia's own DGX Spark mini PC, partly because RAM shortages have pushed component costs up. (The Verge)
Architect launched Liquid Inference, a live auction market for LLM inference requests. Architect Financial Technologies built an LLM router that runs a real-time auction for every prompt: providers bid to serve the request, and the buyer pays the lowest offer that meets its defined rules on latency, model, or cost. The system is designed to let developers set constraints rather than pick a single provider and lock in. It is a different model from static API pricing, closer to how cloud spot instances work. Available now for developers. (MarkTechPost)
The thread
Small models, lower prices, same context
Anthropic's Haiku 5.5 at $0.10 per million input tokens and Liquid AI's open-weight d1 models both push capable inference toward the cheap end of the stack. Haiku 5.5 keeps the 1M context window that used to require a flagship model; d1 skips text generation entirely to cut costs further for classification tasks. Together they make it harder to justify routing every request to a large, expensive model - and Architect's auction-based inference market is built exactly for developers who now have enough options to shop around.
From the publisher
Snippt
Image, video and voice tools in one place
On our radar
- Liquid AI open-sourced two decision models, d1-3B and d1-omni-600M, that read text and images (or audio) and return a typed answer in a single forward pass with zero output tokens - useful for classification and routing tasks where you do not need generated text.
- NVIDIA researchers published PivotOPD, a training method that teaches multi-turn LLM agents to recognize and recover from early mistakes, posting the best average score across 13 baselines on three agent benchmarks.
- A llama.cpp pull request merged a CUDA kernel change that speeds up Qwen 3.x prompt processing by roughly 5.4%, with token generation unchanged - useful if you run large Qwen contexts locally.
- Meta's Muse AI agent app added native iPad support one month after its iOS debut, where it reached the top free app spot on the App Store.
- OpenAI added a College Planner to ChatGPT for Teens, bundling application requirements, deadlines, and notecards into the teen-mode interface.
- A community project called Museum of Models has logged the same prompts sent to 371 models since May 2024, including 93 that no longer exist, all browsable free without a login at museumofmodels.com.
AI worth acting on, in your inbox
Get AI Weekly every Friday: the AI news, tools and tactics that matter for marketers, sellers and creators. Two minutes, with audio.
You can change these choices or unsubscribe at any time. Privacy Policy