Welcome back, fellow AI enthusiast. OpenAI just made GPT-5.6 public. Meta pushed forward with Muse Spark 1.1, its answer to the agentic coding race. We'll also show you how to turn a photo of your fridge into a week of dinners, plus three useful AI tools worth adding to your kit. It's a big one today, let's get into it.
Today In AI:
OpenAI makes GPT-5.6 public
Meta continues with Muse Spark 1.1
Turn a photo of your fridge into a week of dinners
3 useful AI tools and more
Latest News
OPENAI
source: OpenAI
Summary: OpenAI just launched its GPT-5.6 family, moving Sol, Terra, and Luna into general availability, and pairing it with ChatGPT Work and a Codex merger into one desktop app that advances OpenAI's superapp ambitions.
Details:
Sol tops Claude Fable 5 on the Artificial Analysis Coding Agent Index, with major upgrades across agentic coding, computer use, and cybersecurity tasks.
Pricing spans $5/$30 per million tokens for Sol to $1/$6 for Luna, with a new "ultra" mode running four parallel agents on hard tasks.
ChatGPT Work merges with the Codex desktop app, giving OpenAI's answer to Claude Cowork a built-in browser and computer-control agent for everyday tasks.
OpenAI says Sol sets a new high on Agents' Last Exam, beating Fable 5 by over 13 points at a fraction of the compute.
Why it matters: OpenAI didn't just ship a faster model, it shipped a system, folding Codex, ChatGPT Work, and three pricing tiers into one push to become a daily work platform. With Fable and Mythos setting the intelligence bar, GPT-5.6's edge is economics: strong results at a much lower price.
Meta
Summary: Meta just released Muse Spark 1.1, its first model behind a paid developer API, pairing stronger agentic and coding performance with aggressive pricing as Alexandr Wang's Superintelligence Labs races to close the gap with OpenAI and Anthropic.
Details:
Muse Spark 1.1 tops Meta's benchmarks for tool use and orchestration, though it still trails Opus 4.8 and GPT-5.5 on pure coding accuracy.
Developer pricing lands at $1.25 input and $4.25 output per million tokens, positioning it just above Claude Haiku 4.5 and GPT-5.6 Luna.
The model handles a million-token context, generalizes to unfamiliar tools and MCP servers, and can operate a browser or write scripts autonomously.
It's Meta's first paid model API, marking a shift from its open-weight strategy after $145 billion in AI spending under Wang's new lab.
Why it matters: Muse Spark 1.1 won't top coding leaderboards, but Meta didn't build it to, it's an orchestration play, tuned for the agentic work enterprises actually run. Charging for it for the first time also signals Meta now sees its models as a real product, not just an open-weight give-away.
AI Tutorial
Turn a photo of your fridge into a week of dinners
Take a photo of your open fridge and pantry shelf
Tell Claude how many people, and anything to avoid
Ask for 5 dinners using mostly what's already there
Ask for one short grocery list for what's missing
Example Prompt: "Here's a photo of my fridge and pantry. Suggest 5 dinners for 4 people using mostly what I already have. Avoid nuts. Then give me one short grocery list for anything I'm missing."
Quick Hits
📈 Trending AI Tools
Granola — an AI notepad that skips the "bot joined the call" awkwardness, quietly turning your own rough notes into a polished summary
Napkin AI — pastes text in, hands back an instant flowchart or diagram, no design skill or prompting needed
Goblin Tools — free, breaks an overwhelming to-do list into tiny doable steps and rewrites messages by tone
📰 Everything else in AI this week
GPT-5.6 and Grok 4.5 launched within a day of each other, kicking off the most competitive week yet between frontier labs
Security researchers documented JADEPUFFER, the first known fully autonomous AI-driven ransomware attack
Five national cyber agencies (US, UK, Australia, Canada, New Zealand) jointly warned about risks of deploying AI agents in critical infrastructure
Anthropic reportedly overtook OpenAI in annualized revenue for the first time, even as OpenAI keeps the lead on total users