The Allen Institute for AI (Ai2) released a new generation of its flagship large language models, designed to compete more squarely with industry and academic heavyweights.

The Seattle-based nonprofit unveiled Olmo 3, a collection of open language models that it says outperforms fully open models such as Stanford’s Marin and commercial open-weight models like Meta’s Llama 3.1.

Earlier versions of Olmo were framed mainly as scientific tools for understanding how AI models are built. With Olmo 3, Ai2 is expanding its focus, positioning the models as powerful, efficient, and transparent systems suitable for real-world use, including commercial applications.

“Olmo 3 proves that openness and performance can advance together,” said Ali Farhadi, the Ai2 CEO, in a press release Thursday morning announcing the new models.

It’s part of a broader evolution in the AI world. Over the past year, increasingly powerful open models from companies and universities — including Meta, DeepSeek, Qwen, and Stanford — have started to rival the performance of proprietary systems from big tech companies.

Many of the latest open models are designed to show their reasoning step-by-step — commonly called “thinking” models — which has become a key benchmark in the field.

Ai2 is releasing Olmo 3 in multiple versions: Olmo 3 Base (the core foundation model); Olmo 3 Instruct (tuned to follow user directions); Olmo 3 Think (designed to show more explicit reasoning); and Olmo 3 RL Zero (an experimental model trained with reinforcement learning).

Open models have been gaining traction with startups and businesses that want more control over costs and data, along with clearer visibility into how the technology works. 

Ai2 is going further by releasing the full “model flow” behind Olmo 3 — a set of snapshots showing how the model progressed through each stage of training. In addition, an updated OlmoTrace tool will let researchers link a model’s reasoning steps back to the specific data and training decisions that influenced them.

In terms of energy and cost efficiency, Ai2 says the new Olmo base model is 2.5 times more efficient to train than Meta’s Llama 3.1 (based on GPU-hours per token, comparing Olmo 3 Base to Meta’s 8B post-trained model). Much of this gain comes from training Olmo 3 on far fewer tokens than comparable systems, in some cases six times fewer than rival models.

Among other improvements, Ai2 says Olmo 3 can read or analyze much longer documents at once, with support for inputs up to 65,000 tokens, about the length of a short book chapter.

Founded in 2014 by the late Microsoft co-founder Paul Allen, Ai2 has long operated as a research-focused nonprofit, developing open-source tools and models while bigger commercial labs dominated the spotlight. The institute has made a series of moves this year to elevate its profile while preserving its mission of developing AI to solve the world’s biggest problems.

In August, Ai2 was selected by the National Science Foundation and Nvidia for a landmark $152 million initiative to build fully open multimodal AI models for scientific research, positioning the institute to serve as a key contributor to the nation’s AI backbone. 

It also serves as the key technical partner for the Cancer AI Alliance, helping Fred Hutch and other top U.S. cancer centers train AI models on clinical data without exposing patient records.

Olmo 3 is available now on Hugging Face and Ai2’s model playground.

A new platform from the Allen Institute for AI promises to deliver insights into the state of the planet, in near real-time, by giving organizations without deep AI expertise the ability to monitor deforestation, assess crop health, and predict wildfire risk, among other capabilities.

OlmoEarth, announced Tuesday by the Seattle-based nonprofit AI institute, is an open, end-to-end system that uses AI to analyze current and historical satellite and sensor data.

It runs on a new family of AI models, which Ai2 says it trained on millions of Earth observations totaling roughly 10 terabytes of data. The idea is to give anyone free access to the kinds of capabilities typically restricted to proprietary systems or well-resourced AI labs. 

The platform includes tools such as OlmoEarth Studio, a workspace for creating datasets and fine-tuning models, and OlmoEarth Viewer, a web app for exploring AI-generated maps.

The initiative is “making Earth AI accessible to those working on the front lines,” said Ali Farhadi, Ai2 CEO and University of Washington professor, in a press release announcing OlmoEarth.

Patrick Beukema, lead researcher on the OlmoEarth team, added that the project is meant to encourage collaboration across scientific and technical fields, helping different groups work together on shared data and tools to better understand and respond to environmental challenges.

Ai2 said early adopters of OlmoEarth are already showing the potential, using it to update global mangrove maps twice as fast with 97% accuracy, detect deforestation across the Amazon, and map vegetation dryness in Oregon to improve wildfire prediction and prevention, for example.

It’s the latest example of Ai2’s push for “true openness” in AI, extending the philosophy behind its open-weight language and multimodal models into climate science and conservation.

Geospatial analysis has long been dominated by major tech and research organizations. Platforms like Google Earth Engine and Microsoft’s Planetary Computer provide cloud access to petabytes of satellite data, but often require significant technical expertise for analysis and are not fully open-source. 

Ai2 is positioning OlmoEarth as an end-to-end open alternative, providing not just data access but a complete, usable system for model fine-tuning and deployment.

At the model level, Ai2 is competing with AI-specific tools from other major labs. In its research, Ai2 contrasts OlmoEarth with Google’s AlphaEarth Foundations, noting that Google released “embeddings” rather than the open model itself. Ai2 says a fine-tuned OlmoEarth “outperformed AEF substantially,” and also did well against models from Meta, IBM, and NASA.

The OlmoEarth Viewer is available starting today, and Ai2 has released accompanying code and documentation on GitHub. The full platform, including OlmoEarth Studio, is rolling out to select partners, and Ai2 is inviting additional collaborations.

Seattle is looking to celebrate and accelerate its leadership in artificial intelligence at the very moment the first wave of the AI economy is crashing down on the region’s tech workforce.

That contrast was hard to miss Monday evening at the opening reception for Seattle AI Week 2025 at Pier 70. On stage, panels offered a healthy dose of optimism about building the AI future. In the crowd, buzz about Amazon’s impending layoffs brought the reality of the moment back to earth.

A region that rose with Microsoft and then Amazon is now dealing with the consequences of Big Tech’s AI-era restructuring. Companies that hired by the thousands are now thinning their ranks in the name of efficiency and focus — a dose of corporate realism for the local tech economy.

The double-edged nature of this shift is not lost on Washington Gov. Bob Ferguson.

“AI, and the future of AI, and what that means for our state and the world — each day I do this job, the more that moves up in my mind in terms of the challenges and the opportunities we have,” Ferguson told the AI Week crowd. He touted Washington’s concentration of AI jobs, saying his goal is to maximize the benefits of AI while minimizing its downsides.

Seattle AI Week, led by the Washington Technology Industry Association, was started last year after a Forbes list of the nation’s top 50 AI startups included none from Seattle, said the WTIA’s Nick Ellingson, opening this year’s event. That didn’t seem right. Was it a messaging problem?

“A bunch of us got together and said, let’s talk about all the cool things happening around AI in Seattle, and let’s expand the tent beyond just tech things that are happening,” Ellingson explained.

So maybe that’s the best measuring stick: how many startups will this latest shakeout spark, and how can the Seattle region’s startup and tech leaders make it happen? Can the region become less dependent on the whims of the Microsoft and Amazon C-suites in the process? 

“Washington has so much opportunity. It’s one of the few capitals of AI in the world,” said WTIA’s Arry Yu in her opening remarks. “People talk about China, people talk about Silicon Valley — there are a few contenders, but really, it’s here in Seattle. … The future is built on data, on powerful technology, but also on community. That’s what makes this place different.”

And yet, “AI is a sleepy scene in Seattle, where people work at their companies, but there’s very little activity and cross-pollinating outside of this,” said Nathan Lambert, senior research scientist with the Allen Institute for AI, during the opening panel discussion.

No, we don’t want to become San Francisco or Silicon Valley, Lambert added. But that doesn’t mean the region can’t cherry-pick some of the ingredients that put Bay Area tech on top.

Whether laid-off tech workers will start their own companies is a common question after layoffs like this. In the Seattle region at least, that outcome has been more fantasy than reality. 

This is where AI could change things, if not with the fabled one-person unicorn then with a bigger wave of new companies born of this employment downturn. Who knows, maybe one will even land on that elusive Forbes AI 50 list. (Hey, a region can dream!)

But as the new AI reality unfolds in the regional workforce, maybe the best question to ask is whether Seattle’s next big thing can come from its own backyard again.

Related: Ferguson’s AI balancing act: Washington governor wants to harness innovation while minimizing harms