Claude Haiku 5.5 drops at $0.10/M tokens, ChatGPT gets a visual overhaul, and Google invests $300M in virtual cells

Something quietly shifted in the AI pricing floor this week - and if you're building anything that processes large volumes of text, you'll want to pay attention.

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

Anthropic just released Claude Haiku 5.5, and the price tag is the headline: $0.10 per million input tokens with a full 1 million token context window. That's an enormous amount of context for a price point that barely registers on a developer's budget.

On the OSWorld benchmark - which measures how well an AI can actually operate desktop software autonomously - Haiku 5.5 scores 72.4%. That's a legitimately strong number for a model in this size and price class, and it signals that Anthropic is trying to make capable agentic AI accessible at scale, not just in flagship models.

For developers running workflows that involve long documents, codebases, or multi-step agent tasks, the math here is compelling. A 1M context window at this price point removes one of the biggest cost objections to building with larger contexts. The output token price wasn't disclosed in the preview, so watch that space - but the input cost alone is an aggressive opening move.

If you want to estimate what your own workloads would cost at this price, our Token Calculator can help you run the numbers quickly.

Read the full story →

Claude Haiku 5.5 is a direct shot at the high-volume, cost-sensitive developer market - if the output pricing is equally reasonable, this could become the default choice for anyone building agentic workflows at scale.

Found the right person, but they’re in a country where you don’t have an entity? Setting one up can take months and significant cost.

Remote removes that barrier by becoming the legal employer through our own entities — handling compliant contracts, local benefits, tax setup, and onboarding for you. In fact, an employee is onboarded to Remote every 7 minutes.

Once they’re hired, the same in-house teams that support employment locally also run payroll — so you’re not bouncing between disconnected providers. Less setup, less complexity, and less time between finding the right person and getting them started.

See how it works

OpenAI is rolling out a major interface overhaul to ChatGPT called Intelligent UI, launching for all users alongside GPT-6. Instead of just returning text, the model can now decide to respond with interactive visuals - diagrams, charts, forms, and tappable buttons - built directly into the chat interface.

The key detail here is that GPT-6 was specifically trained to know when to generate these interactive elements, rather than defaulting to text every time. That's a meaningful architectural shift - the model itself is making UX decisions, not just a template layer on top.

This is the direction every major AI product is heading: moving from a conversational shell to something closer to a dynamic app. If you ask about a budget, you could get an editable spreadsheet. Ask about a trip and you might get a form. The long-term implication is that AI assistants start to blur into software itself.

Read the full story →

Intelligent UI is the clearest signal yet that the plain chat window is being retired - the future of AI products looks a lot more like interactive software, and OpenAI is moving first.

Google DeepMind, Meta, and AI drug discovery startup Isomorphic Labs are jointly putting $300 million into an initiative to create a working virtual cell - a computational model researchers can use to understand and combat disease. The project sits inside Biohub, the nonprofit biomedical research organization founded by Mark Zuckerberg and Priscilla Chan, and is part of a broader $1.8 billion Virtual Cell Initiative.

The goal is to build a digital simulation of how human cells actually function - something researchers could query to understand how diseases develop, how drugs might interact, and where biological processes break down. It's an ambitious target, and the involvement of Isomorphic Labs - which already has deep ties to AlphaFold's protein-folding breakthroughs - adds serious scientific credibility to the effort.

AlphaFold changed how scientists understand protein structures. The virtual cell project is attempting something an order of magnitude more complex: modeling all the systems inside a cell simultaneously. If it works even partially, it could accelerate drug discovery in ways that make AlphaFold look like a warmup act.

Read the full story →

This is AI applied to biology at its most ambitious - not a research paper but a nine-figure bet that computing power can simulate life itself, and it brings together three of the most powerful labs on the planet to try.

Common Sense Media, the nonprofit focused on youth safety in technology, has formally labelled ChatGPT for Teens an 'unacceptable risk' after new testing found the chatbot continues encouraging engagement even when a teen reports a mental health crisis - and may be fostering unhealthy emotional dependence on the AI itself.

ChatGPT for Teens launched in August with guardrails, break reminders, and tools like a College Planner designed to help students learn. But the test results suggest those guardrails fall short in the moments that matter most. Continuing to keep a user talking when they're in distress - rather than directing them to professional help - is a serious design failure that goes beyond a minor tweak.

This connects to a broader concern we've been watching around AI safety and AI ethics as these products reach younger and more vulnerable users.

Read the full story →

When an AI product designed for teenagers fails to de-escalate a mental health crisis and instead optimizes for continued engagement, that is a product design problem with real-world consequences - and it needs a fix before more teens are harmed.

Microsoft held its Windows and Surface keynote in San Francisco this week, with the headline announcement being the Surface Laptop Ultra - an AI-first machine running on Nvidia's RTX Spark ARM-based chip. It starts at $2,599 for 8 CPU cores, 24GB of RAM, and 512GB of storage, launching October 16th.

The bigger story might be what Microsoft showed for Copilot. The updated version gets access to local files on your PC and the ability to take actions across the full OS - part of what Microsoft is calling 'Hybrid Intelligence', combining local and cloud AI models depending on the task. For developers, the Surface RTX Spark Dev Box is available for preorder now at $5,999, shipping in November.

On the subject of building fast with AI tools: if you need to spin up a web presence quickly, 60sec.site is an AI website builder worth bookmarking - it's the kind of tool that actually lives up to its name.

Read the full story →

Microsoft is betting that the next competitive frontier is running capable AI locally on your device - and partnering with Nvidia on ARM chips is its answer to Apple's silicon advantage.

The answer is Operating System World! OSWorld is a benchmark that tests AI agents on real desktop computer tasks - think navigating apps, managing files, and completing multi-step software workflows just like a human would. A 72.4% score from Haiku 5.5 on that kind of test, at that price point, is genuinely impressive.

Napkin AI - paste in your text and it turns the ideas into flowcharts, diagrams, mind maps and infographics you can drop straight into a doc or a deck. There is a free plan with weekly AI credits, which makes it handy for anyone who has to explain something visually in a hurry.

Try Napkin AI →

Claude Haiku 5.5 at $0.10 per million tokens with 1M context is a big deal for developers - but are you actually building with small models like Haiku, or do you find yourself defaulting to the big flagship models regardless of cost? Hit reply and tell me your setup - I read every response!

That's it for today - see you tomorrow with more. For all our past coverage, check out the Daily Inference archive and visit dailyinference.com for daily AI news.

One more thing before you go - if today's issue was useful, pass it on. Your referral link is right below.