SUBSCRIBE | ADVERTISE

Mid-December AI news sans hype

OpenAI went from code red to the launch of GPT-5.2 and a Disney deal

Was this email forwarded to you? Sign up here.

Howdy wizards,

OpenAI dominated the AI news landscape the first two weeks of December. After declaring an internal code red (following the Gemini 3 Pro launch), they’ve bounced back with a new model which delivers the goods and a major collaboration with Disney.

Here’s what’s brewing in AI.

Models

• OpenAI launched GPT-5.2, their newest flagship model. It exceeds GPT-5.1 on all key benchmarks and might be the world’s best model so far at end-to-end tasks since it combines exceptional long-context understanding, agentic tool calling and vision. ChatGPT’s ability to create sophisticated spreadsheets and presentations is much better with this upgrade.

  • The release probably came faster than originally planned, after the narrative of Gemini 3 Pro being way better than ChatGPT had the company declaring a code red internally (e.g. had people saying OpenAI is going to be the next Netscape). With this new release, it’s shaping up to be quite the turn-around, though. I’ve been using the new model for long-running coding tasks this week and can confirm, OAI is back again.

• DeepSeek launched V3.2 and V3.2‑Speciale and claims GPT‑5 / Gemini 3 Pro–level performance. The 685B-parameter releases are under an MIT license and runs at a fraction of the cost of GPT-5.2 ($0.28/$0.42 per 1 million input/output tokens). V3.2‑Speciale is a model version specifically for deep reasoning and is frontier-level at math.

• Google rolled out Gemini 3 Deep Think in the Gemini app. It can explore multiple hypotheses simultaneously and builds on Gemini 2.5 Deep Think, which won an IMO gold medal.

• Mistral launched three things: Mistral 3, Devstral 2 + Vibe CLI. The first is a lineup of four new open source models: 3B, 8B, 14B, and the big one, Mistral Large 3 (675B parameters). Devstral 2 is a coding model hitting a respectable 72.2% on SWE-bench. While models are still behind the big labs (OAI, Anthropic, Google) performance wise, it will be very interesting to see what happens if/when they reach performance parity as Apache 2.0 licensing is very attractive to most orgs.

IN PARTNERSHIP WITH GRANOLA

AI meeting notes you'll actually use

You know which points matter most in your meetings.

Granola lets you capture them, then uses AI to turn your notes into summaries and action items you'll actually use.

Try Granola free →

Industry moves

• Disney invested $1B in OpenAI and became Sora’s first major content partner. This seems like a strong business move by OAI; in a time that they’re being sued by major publishers like NYT for plagiarism, they’re buying legal clearance to flood the internet with more than 200 Disney, Marvel, Pixar and Star Wars characters. They no longer have to pretend like they didn’t train on Disney content.

• OpenAI, Anthropic, and Block launched the Agentic AI Foundation under the Linux Foundation to standardize agent protocols. The group, whose mission is to ensure agentic AI evolves transparently and collaboratively, says it will steward projects like Anthropic’s MCP (Model Context Protocol), OpenAI’s AGENTS.md, and Block’s Goose framework. Supporting members including Google, Microsoft, AWS, and Cloudflare.

• Anthropic acquired Bun. Bun is a JavaScript runtime and toolchain used for fast development workflows. Claude Code is now a billion dollar product, and it’s easier for Anthropic to just buy the infrastructure they need rather than to build and harden it in public.

New tools & product features

• Anthropic launched a beta Claude Code for Slack integration that turns a Slack thread into a coding task for Claude. Tagging @Claude spins up a Claude Code session that can select repos, post progress back to Slack, and deliver links to review changes/open PRs. Slack keeps expanding into the UI layer for teamwork (tickets, approvals, deploys) and now code changes.

• Google launched Workspace Studio to build agents with natural-language commands across Gmail, Drive, and other Workspace apps. The idea is to enable automation of routine workflows without stitching together multiple tools. Useful to remember in this context of where automating work gets easier: AI-powered workflows are non-deterministic. You’ll get occasional variances/errors; think through if and where it makes sense to have a human in the loop.

• Runway released Gen-4.5 improving realism, motion, and control (including more consistent details like hair/fabric across frames).

Research

As we’re nearing end of year, three major players released detailed reports of how people and companies are using their platforms:

• OpenAI published its first State of Enterprise AI report, summarizing usage data from 1M+ workplace accounts and a survey of 9,000 workers. ChatGPT Business users save ~40–60 minutes/day on average and 75% of workers report doing tasks they “couldn’t before.”

• Anthropic introduced Anthropic Interviewer, a Claude-powered tool for running and analyzing qualitative interviews at scale (basically a guided questionnaire plus voice). They’ve used it to summarize interviews with 1,250 professionals; 86% of report that AI saves them time and 69% say they face some kind of social stigma around using AI.

• Microsoft published the Copilot Usage Report 2025, analyzing 37M+ Copilot conversations to map what people do with AI. Health questions were the top category across devices and common uses included search, advice, creation, learning, and technical support.

What I’m actually using

• I’ve started using Opus 4.5 and GPT-5.2 for coding this week and they feel equally amazing—I can’t tell which one is best. I use both inside Cursor as IDE extensions. I’ve been testing especially GPT-5.2 on some long-running tasks (hours) and I’m a bit blown away; it’s actually way better than Codex-Max which came out not long ago. It’s pretty strange, every time I feel like I’ve hit a wall in my vibe coding journey, new models come along that just solve all my current sticking points. In the case of Opus 4.5, the main drawback for me is that the usage caps are very limited on the Pro plan; I keep hitting walls after 3-4 prompts and need to wait for 4-5 hours until I can start using it again.

  • As as result of these two new models being so good, I’ve stopped using Google’s Antigravity IDE this week because I can’t use the IDE extensions mentioned above inside it, and Gemini 3 Pro is nowhere near the same level for me.

What’s on my radar

• I haven’t tested GPT-5.2 for tasks other than coding yet, but I’ll definitely experiment with it more over the next weeks.

• I think it’s really cool that Mistral is European and that they’re doing open-source coding models and CLI tools now. But I’m not going to start using them for coding until they are truly on par with the other models. It might be viable for seasoned developers using these models mostly for code completion, but for purebred vibe coders (like me), I’ve found that the so-called incremental improvements to model performance really makes a massive difference to what I can achieve. So it makes sense to stick with products from the big labs for now.

FEATURED GPT

A calm space to slow down, understand your emotions, and make relationship decisions you can stand behind—without rushing or avoiding what you feel.

Check it out here

Want your custom GPT featured? Apply here.

I hope you’re having a wonderful weekend.

If you have an extra 32 seconds to spare, I’d be delighted if you’d let me know how you’re liking the newsletter and how I can improve. Just hit reply or leave a comment in the poll below.

ALL RIGHT, ENOUGH AI FOR TODAY

Disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes work with sponsors and may earn a commission if you buy something through a link in here. If you choose to click, subscribe, or buy through any of them, THANK YOU – it will make it possible for me to continue to do this.