ngl ChatGPT wrapped was pretty on point Happy holidays all!

80 posts this year on Interconnects, my most ever, and likely a local maximum. A short reflection on what happens when your two long-term areas of focus — open models and RL — become the centerpiece of the AI boom. Thx for your support! www.interconnects.ai/p/2025-inter...

I'm making big updates to the RLHF book in the coming weeks, so if you've been reading it & have feedback, please open a GitHub issue. Or, if you took a look at only the initial versions, look again as its come a long way with a trickle of improvements. 🍒 rlhfbook dot com

Can't believe a product called Nano Banana is what is going to take plucky underdog Google Gemini to finally beat the incumbent ChatGPT. Fast, precise image gen and editing is such a joy to use and surely mass market adoption is cranking. More warranted to call a Code Red for that than Gemini 3.

A very nice long read on Olmo 3 from my friend Cameron. He puts in so much care to do a great job on these. cameronrwolfe.substack.com/p/olmo-3

Open models year in review. What a year! We're back with an updated open model builder tier list, our top models of the year, and our predictions for 2026. www.interconnects.ai/p/2025-open-...

Too many cases of starting something in Codex that you think is going great then Opus needs to save the day.

Olmo 3.1: even more RL = even more RL-Zero! @saurabhshah2.bsky.social and I tweaked some hyperparams and prompts, @hamishivi.bsky.social and @finbarr.bsky.social improved the code and boom! New Olmo 3.1 RL-Zero 👾 An updated, solid baseline for your RL and reasoning research

My favorite of the Olmo 3 paper: Transparent auditing of the cost for the v3 models (not 3.1), based of wall clock time for pre/post train, evals, cluster issues, etc, as a counter to the famous $5.576M for DeepSeek V3 at $2/H100 hour, Olmo 3 start to end would cost $2.75M allenai.org/papers/olmo3

Olmo 3.1 32B Think shows that not just frontier labs can scale RL. My favorite RL run yet over 7+ years of doing RL. The biggest fully open RL run ever? Gold stars on downstream evals is our original release, this latest one is the final checkpoint on the plot.

Remember when OpenAI said GPT-5 was when they were going to fix their naming scheme.

It's finally here: The public (and most complete) version of my talk covering every stage of the process to build Olmo 3 Think. This involves changes and new considerations of every angle of the stack, from pretraining, evaluation, and of course post-training.

Building Olmo 3 Think Foundations of Reasoning in Language Models @ NeurIPS 2025 Today 13:45 - 14:30

Good researchers obsess over evals The story of Olmo 3 (post-training), told through evals NeurIPS Talk tomorrow. Upper Level Room 2, 10:35AM. Slides: docs.google.com/presentation...

Loved seeing our OSU collaborator Sam Stevens featured at the #NeurIPS SAIL Media Booth w/ @natolambert.bsky.social! Great convo on how AI accelerates scientific discovery (not just bigger models 🙃). Clips ⬇️ youtube.com/shorts/v0k5D... youtube.com/shorts/4R-BF... #AI4Science #NeurIPS2025

On a prompt count basis this mean reasoning models are not close to a majority on OpenRouter, as reasoning models can use 10-1000x the tokens of non-thinking models per prompt. Lots of need for fast, efficient open models. Reasoning model usage is likely closed labs more.

Olmo 3 32B Think is now free on Open Router until 12/22. Let us know what you think and what to improve! (Hosted by Parasail) This may give it the hug of death... would be my dream. openrouter.ai/allenai/olmo...

Where to find me at NeurIPS! Talks, panels, socials Talks: I'm giving two talks on the last day at Workshops Dec. 7th 1. Good researchers obsess over evals. 10:35am-11:05am: Evaluating the Evolving LLM Lifecycle 2. Building Olmo 3 Think. 1:45pm-2:30pm: Foundations of Reasoning in Language Models

Excited to share another NeurIPS event I'm helping with. We're hosting a dedicated booth to record researchers talking about their work, share that audio&video content on our socials, and start great conversations. calendly.com/contact-read...

New Artificial Analysis Openness Index! As it should be, Olmo is king. This plot feels like a solid representation of ability v openness. My favorite part is Olmo 3 mogging on Llama 4 Maverick in every plot 🤭. It's pretty sad that the two values are pretty much perfectly anti-correlated.

Does anyone use the Deep Research features anymore. What's SOTA? My guy would guess Gemini 3 is decent? I've just been so GPT Pro pilled. Why use DR?

Olmo 3 is a banger paper. Finally getting to read the rest of it 😅. We love releasing things that serve as a comprehensive snapshot of public knowledge on training leading language models. There's an award for whoever finds all the secrets first in the new arxiv version allenai.org/papers/olmo3

Opus? Sorry, living under rocks today.

Latest open artifacts (#16): Who's building models in the U.S., China's model release playbook, and a resurgence of truly open models A month with SOTA releases with (truly) open model releases left and right. www.interconnects.ai/p/latest-ope...

I asked (on ChinaTalk) the head of product at Z ai, one of the leading Chinese companies building open models, how long it takes them to get their model out the door once its done training. Incredible stuff: "a few hours" and the model is on HuggingFace. www.chinatalk.media/p/the-zai-pl...

People: "You must be so relaxed, proud, and happy that the model you worked on all year is out." Me:

the Epstein files have been trending on HuggingFace. > This dataset is provided for... > Evaluating information retrieval and retrieval augmented generation (RAG) systems. > It is not intended for: Fine-tuning language models. ??

Happy Olmo day to all who celebrate. Sorry to all who delayed releases today to get out of our way. We're hiring.

We present Olmo 3, our next family of fully open, leading language models. This family of 7B and 32B models represents: 1. The best 32B base model. 2. The best 7B Western thinking & instruct models. 3. The first 32B (or larger) fully open reasoning model.

Chinese models are enabling AI research. US progress needs to be accelerated

Conspiracy theory: OpenAI's serving costs were too high because GPT 5 Thinking overthought too much on easy queries, and GPT 5.1 was a patch to fix that. 5.1 just feels weird, can't quite place it.

A new tab on Google Scholar??? scholar.google.com/scholar_labs...

Why AI writing is mid How the current way of training language models destroys any voice (and hope of good writing). www.interconnects.ai/p/why-ai-wri...

Updating to GPT 5.1 in codex made the model do crazy shit when I ask it to stage changes, make a PR, revert to XYZ branch, etc. Very shocking because even the first claude code could do this. Am I the only one?

GPT 5.1 Thinking sometimes responds so much faster than GPT 5 Thinking would that it makes me think "shit I must've used the wrong model" and doubt it's answer a bit. Funny dynamic. TBD if quality dropped at all, quality at least didn't nosedive. Hard to tell when style changes a bit.

I'm excited to announce my RLHF Book is now in pre-order for the @manning.com Early Access Program (MEAP), and for this milestone it's 50% off. Excited to land in print in early 2026! Lots of improvements coming soon. Thanks for the support! hubs.la/Q03Tc37Q0

Many people are sleeping on, or even making fun of this plot in the GPT 5.1 release. This is a crucial plot for anyone serving a thinking model in real world use-cases. Latency to an answer is a huge cause of user churn and not thinking enough is a fast track to having your model's output be bad.

OpenAI showing very clearly why you should care about Character Training with GPT 5.1: It's the leading selling point of the release.

Lol currently every X account with 2fa auth enabled is locked out, while if you don't have 2fa you can use the app as usual. Iconic levels of broken.

I’m starting a new series of interviews with all the leading open model labs around the world to show why people are doing this, how people train great models, and where the ecosystem is going.

If you're working on character training research, what're you working on? What is limiting your ability to do the research you want here? Surely there are more people studying how to modify & steer model personality after the GPT 4o sycophancy incident.

Opening the black box of character training Some new research from me! Exploring how easy it is to craft personalities like sycophantic chatbots, and exploring how this will change as we move from chat to agents. www.interconnects.ai/p/opening-th...

New bike day!

The DeepSeek moment underestimated the talent portion & overfocused on training capital in N of GPUs. The rest of 2025 has been living through that reality with Kimi, GLM, Ant Ling, Meituan... The burden of proof is back on scaling if AI will be in the hands of a few companies.

I appreciate the shoutout from @simonwillison.net I'm building up a much richer (and direct) understanding of Chinese AI labs. Excited to share more here soon :)

The Chinese Kimi K2 thinking model beats GPT and Claude on some benchmarks. This analysis from @natolambert.bsky.social is a good overview iew of what is going on www.interconnects.ai/p/kimi-k2-th...

Thoughts on Kimi K2 Thinking Congrats to the Moonshot AI team on the awesome open release. For close followers of Chinese AI models, this isn't shocking, but more inflection points are coming. Pressure is building on US labs with more expensive models. www.interconnects.ai/p/kimi-k2-th...

The Great Lock In

We're starting to hire for our 2026 Olmo interns! Looking for excellent students to do research to help build our best models (primarily enrolled in Ph.D. with experience or interest in any area of the language modeling pipeline). job-boards.greenhouse.io/thealleninst...

The PyTorch recording of my Open Models Recap talk is out. I think this a great and very timely talk, I'm very happy with it and recommend you watch it more than I'd recommend my usual content. (Thanks again to the PyTorch team -- great event) youtu.be/WfwtvzouZGA