🆕 New in Asta: multi-turn report generation. You can now have back-and-forth conversations with Asta, our agentic platform for scientific research, to refine long-form, fully cited reports instead of relying on single-shot prompts.

Now you can use our most powerful models via API. Olmo 3.1 32B Think, our reasoning model for complex problems, is on @openrouter.bsky.social—free through 12/22. And Olmo 3.1 32B Instruct, our flagship chat model with tool use, is available through @hf.co Inference Providers. 👇

🎥 Introducing SAGE, an agentic system for long video reasoning on entertainment videos—sports, vlogs, & more. It learns when to skim, zoom in, & answer questions directly. On our SAGE-Bench eval, SAGE with a Molmo 2 (8B)-based orchestrator lifts accuracy from 61.8% → 66.1%. 🧵

🎗️ Reminder, our Molmo 2 and Olmo 3 Reddit AMA begins soon at 1pm PST / 4pm EST. www.reddit.com/r/LocalLLaMA...

Last year Molmo set SOTA on image benchmarks + pioneered image pointing. Millions of downloads later, Molmo 2 brings Molmo’s grounded multimodal capabilities to video 🎥—and leads many open models on challenging industry video benchmarks. 🧵

🗓️ Tue Dec 16, 1–2pm PT: AMA with researchers + engineers from our Olmo & Molmo teams, hosted by r/LocalLLaMA. 💬 Ask your questions now—we’ll start answering when the AMA begins!

Introducing Bolmo, a new family of byte-level language models built by "byteifying" our open Olmo 3—and to our knowledge, the first fully open byte-level LM to match or surpass SOTA subword models across a wide range of tasks. 🧵

🧠 Introducing NeuroDiscoveryBench. Built with @alleninstitute.org, it’s the first benchmark for evaluating AI systems like our Asta DataVoyager agent on neuroscience data. The benchmark tests whether AI can truly extract insights from complex brain datasets.

Olmo 3.1 is here. We extended our strongest RL run and scaled our instruct recipe to 32B—releasing Olmo 3.1 Think 32B & Olmo 3.1 Instruct 32B, our most capable models yet. 🧵

Update: DataVoyager, which we launched in Preview early this fall, is now available in Asta. 🎉 You can upload real datasets, ask complex research questions in natural language, & get back reproducible answers + visualizations. 🔍📊

We're at #NeurIPS2025 with papers, posters, workshops, fireside chats, & talks across the conference. Come learn about our latest research + see live demos!

SciArena leaderboard update! 🔬 We've added new frontier models – including GPT-5.1 and Gemini 3 Pro Preview – to our arena for scientific literature tasks. The new rankings: o3 holds #1, Gemini 3 Pro Preview lands at #2, Claude Opus 4.1 sits at #3, GPT-5 at #4, & GPT-5.1 debuts at #5. 🧵

Olmo 3 is now available through @hf.co Inference Providers, thanks to Public AI! 🎉 This means you can run our fully open 7B and 32B models — including Think and Instruct variants — via serverless API with no infrastructure to manage.

We recently released OlmoEarth—an open spatio-temporal foundation model for planetary intelligence, trained on multimodal Earth observation, OpenStreetMap, & other open data. It sets a new accuracy/efficiency Pareto frontier and powers our OlmoEarth Platform. Technical report now on arXiv. 👇

⚠️ Update on Deep Research Tulu (DR Tulu), our post-training recipe for deep research agents: we’re releasing an upgraded version of our example agent, DR Tulu-8B (RL), that matches or beats systems like Gemini 3 Pro & Tongyi DeepResearch-30B-A3B on core benchmarks. 🧵

Our Olmo 3 models are now available via API on @openrouter.bsky.social. Try Olmo 3-Instruct (7B) for chat & tool use, and our reasoning models Olmo-3 Think (7B & 32B) for more complex problems.

Missed our Olmo 3 livestream? Check it out on YouTube! The Olmo team recaps what makes Olmo 3 our most exciting LM release yet, perhaps most importantly our fully open model flow of checkpoints, training data & recipes, and more. Watch 📷 www.youtube.com/watch?v=QUFK...

Announcing Olmo 3, a leading fully open LM suite built for reasoning, chat, & tool use, and an open model flow—not just the final weights, but the entire training journey. Best fully open 32B reasoning model & best 32B base model. 🧵

Today we’re releasing Deep Research Tulu (DR Tulu)—the first fully open, end-to-end recipe for long-form deep research, plus an 8B agent you can use right away. Train agents that plan, search, synthesize, & cite across sources, making expert research more accessible. 🧭📚

Introducing OlmoEarth 🌍, state-of-the-art AI foundation models paired with ready-to-use open infrastructure to turn Earth data into clear, up-to-date insights within hours—not years.

Our Olmo Discord AMA just wrapped! Researchers answered community questions about their work with Olmo, our family of fully open LLMs. Here’s some highlights. 🧵

⚠️ Remin​​der: Join us today in our Discord @ 8:00 a.m. PT!

Join us for a live Discord AMA Tues, Oct 28 @ 8:00 a.m. PT with researchers who’ve run real studies using our Olmo language model family—from machine unlearning to knowledge cutoffs to how models acquire new skills. 👇

Our fully open Olmo models enable rigorous, reproducible science—from unlearning to clinical NLP, math learning, & fresher knowledge. Here’s how the research community has leveraged Olmo to make the entire AI ecosystem better + more transparent for all. 🧵

Curious to see how olmOCR 2, our new tool to turn digitized documents into trustworthy, structured text, performs? Check out these examples. 👇

We’re updating olmOCR, our model for turning PDFs & scans into clean text with support for tables, equations, handwriting, & more. olmOCR 2 uses synthetic data + unit tests as verifiable rewards to reach state-of-the-art performance on challenging documents. 🧵

📣 Heads up, Bay Area folks: Research Scientists Nathan Lambert (@natolambert.bsky.social) & Sewon Min (@sewonm.bsky.social) will be giving separate talks during #OpenSourceAIWeek and #PyTorchCon in SF.

🌍 Announcing SamudrACE, our AI climate emulator built so scientists & planners can run “what-if” climate experiments quickly. Traditional models are slow and costly; SamudrACE makes high-quality simulations fast & more accessible. 🧵

We were honored to join the #IUCNcongress 2025 in Abu Dhabi last week. Our message: AI should serve science & society—built openly, deployed with partners, and measured by real-world impact.

📊 Today we're releasing data showing which scientific papers our AI research tool Asta cites most frequently. Think of it as creating citation counts for the AI era—tracking which research is actually powering AI answers across thousands of queries. 🧵

"We are committed to our fully open ethos. That's why we release everything—weights, code, training data, checkpoints, all of it." — @nlpnoah.bsky.social at the Madrona IA Summit last week.

As part of #SeattleAIWeek, we're hosting "AI Innovation in the Open" on Oct. 30 from 2-4:30pm—an afternoon of live demos and hands-on tutorials at Ai2 HQ. 👇

Introducing Asta DataVoyager—our new AI capability in Asta that turns structured data into transparent, reproducible insights. Built for scientists, grounded in open, inspectable workflows. 🧵

A few new challengers enter SciArena—including DeepSeek-V3.2-Exp and Claude Sonnet 4.5 🔬

📢 New #COLM2025 paper 📢 Standard benchmarks give every LLM the same questions. This is like testing 5th graders and college seniors with *one* exam! 🥴 Meet Fluid Benchmarking, a capability-adaptive eval method delivering lower variance, higher validity, and reduced cost. 🧵

🚀 Introducing Fluid Benchmarking—an adaptive way to evaluate LLMs. Inspired by psychometrics, it tailors which questions to ask based on each model’s capability, making evals more efficient & reliable. 🧵

📓 New from Ai2: we’ve released source code that shows how we built AskOlmo, our Discord chatbot powered by our Olmo model family. It’s a peek behind the curtain—so you can see how it all came together. 👇

🚀 New in the Ai2 Playground: side-by-side comparison is live. Compare two Ai2 models with the same prompt and see the results next to each other. ⚖️🆚

🌍☀️❄️ Can AI forecast year-to-year differences in the seasons? New research with @metoffice.gov.uk shows our ACE2 ML model demonstrates seasonal forecasting skill—matching traditional physics-based methods while using dramatically less compute. 🧵

✨Meet OLMoASR✨ By pairing our curated 1M-hour dataset with a powerful architecture, we've built open ASR models that achieve competitive performance with models like Whisper. We're open-sourcing data, code and models to help the community build more robust and transparent ASR.

🎙️ Say hello to OLMoASR—our fully open, from-scratch speech-to-text (STT) model. Trained on a curated audio-text set, it boosts zero-shot ASR and now powers STT in the Ai2 Playground. 👇

We're honored to be named a @fastcompany.com Corporate Wellness Innovator 2025 (Tech)! 🏆 Our Seattle HQ was built for well-being with fresh air, outdoor spaces, and movement-friendly designs to help our team thrive.

Today we’re releasing agent-baselines, a suite of 22 classes of AI agents for science—including 9 open-source research-tuned agents like our state-of-the-art, benchmark-leading Asta v0. 🚀🔬 Part of our Asta ecosystem to advance scientific AI. 👇

As part of Asta, our initiative to accelerate science with trustworthy AI agents, we built AstaBench—the first comprehensive benchmark to compare them. ⚖️

Introducing Asta—our bold initiative to accelerate science with trustworthy, capable agents, benchmarks, & developer resources that bring clarity to the landscape of scientific AI + agents. 🧵

🚨 SciArena update + evaluation of new models including GPT-5! 🚨 With thousands of new votes, new LLMs are reshaping our leaderboard for scientific literature tasks. o3 still leads—but GPT-5, Claude Opus 4.1, & more are closing the gap.

🚀 In March, we launched Paper Finder, an LLM-powered literature search agent that surfaces papers other tools miss. Now, we’re releasing an open-source snapshot to enable others to inspect & build on it—and reproduce the results. 🧵

Evaluating language models is tricky, how do we know if our results are real, or due to random chance? We find an answer with two simple metrics: signal, a benchmark’s ability to separate models, and noise, a benchmark’s random variability between training steps 🧵

📢 New paper from Ai2: Signal & Noise asks a simple question—can language model benchmarks detect a true difference in model performance? 🧵

We’re releasing early pre-training checkpoints for OLMo-2-1B to help study how LLM capabilities emerge. They’re fine-grained snapshots intended for analysis, reproduction, and comparison. 🧵