MIT Technology Review · Sep 25, 2024
MIT Technology Review reports Ai2’s case for open-source Molmo models
Markdown · JSON · Original source
MIT Technology Review published this report on September 25, 2024. The Allen Institute for Artificial Intelligence () says its open-source language models, called Molmo, perform as well as top proprietary models on tests involving images, charts and documents. Ai2 attributes the results to carefully curated training data and detailed human descriptions of images, converted from speech into data. The report also describes Molmo’s ability to “point” to image elements by identifying relevant pixels; in a demonstration, the model counted deck chairs but could not locate a specific parking lot. Ai2 CEO Ali Farhadi argues that open-source models let others build applications and can use money and efficiently. ’s Yacine Jernite says Molmo’s significance will depend on what developers build and how people improve it. Farhadi also argues that open-source models can bring returns on AI investment where large, expensive proprietary models will not. These are the speakers’ claims, not established outcomes. Jernite says techniques for producing higher-quality training data could help govern data used in AI development. The report does not give the full test methods or results, so how Molmo’s performance compares beyond the described tests remains unknown.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
MIT Technology Review · Article
Source details
Original source date: Sep 25, 2024 (2024-09-25T13:00:00Z).
Date basis: Date from the publisher’s own records, through its API.
Added to Escape Meridian: 2026-11-17T02:18:24.203Z. First observed: 2026-10-30T06:53:59.581Z. Retrieved: 2026-10-30T06:55:46.192Z.
Referenced by
Recorded citations in public readings. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 30, 2026; originally published September 25, 2024. The report describes performance tests but does not give their full methods or results, limiting assessment of the comparison. The report’s described demonstration includes a failure to locate a specific parking lot; it does not establish broader real-world reliability.
Open question
How will applications built on Molmo, and the ways people improve it, shape its significance beyond the reported tests?
Source-stated
MIT Technology Review reports that Ai2 says Molmo performs as well as top proprietary models on tests involving images, charts and documents.
The Allen Institute for Artificial Intelligence (Ai2), a research nonprofit, is releasing a family of open-source multimodal language models, called Molmo, that it says perform as well as top proprietary models from OpenAI, Google, and Anthropic.
Source-stated
Ai2 attributes Molmo’s performance with fewer resources to a smaller, curated image dataset and detailed human descriptions converted from speech into training data.
Ai2’s Molmo models have been trained on a significantly smaller and more curated data set containing only 600,000 images
Source-stated
In a demonstration described by MIT Technology Review, Molmo counted deck chairs but could not locate a specific parking lot.
It was not perfect, however. It could not locate a specific parking lot, for example.
Source-stated
Ai2 CEO Ali Farhadi argues that open-source models can use money and time efficiently and can bring returns on AI investment where large, expensive proprietary models will not.
Big, expensive proprietary models won’t do that, he argues, but open-source ones can. He says the work shows that open-source AI can also be built in a way that makes efficient use of money and time.
Source-stated
Hugging Face’s Yacine Jernite says Molmo’s significance will depend on the applications developers build and the ways people improve it.
its real significance will lie in the applications developers build on top of it, and the ways people improve it.
Source and contributor context
MIT Technology Review · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Multimodal model
A model that works with more than one kind of input or output, such as text, images and audio.
Ai2
Ai2 develops AI research tools and models, including the openly documented Olmo family.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Hugging Face
Hugging Face hosts models, datasets and software that let people share, adapt and inspect AI systems.
Editorial provenance
Version 1. Publication ID: dca0a5f7-d96e-b864-364f-36382ab6b824.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-17T02:13:56.992Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: d03c408de497422f408f394c63ab260fdab90daf489b4af6ab1a3d4f344d9540. Source text hash: d27367694081826976861f539bdb7db6f93750bf49e81fb3251b07f1f4ce1c29.
Revision and source-check history
Version 1 · 2026-11-17T02:18:24.203Z
MIT Technology Review reports Ai2’s case for open-source Molmo models
MIT Technology Review published this report on September 25, 2024. The Allen Institute for Artificial Intelligence () says its open-source language models, called Molmo, perform as well as top proprietary models on tests involving images, charts and documents. Ai2 attributes the results to carefully curated training data and detailed human descriptions of images, converted from speech into data. The report also describes Molmo’s ability to “point” to image elements by identifying relevant pixels; in a demonstration, the model counted deck chairs but could not locate a specific parking lot. Ai2 CEO Ali Farhadi argues that open-source models let others build applications and can use money and efficiently. ’s Yacine Jernite says Molmo’s significance will depend on what developers build and how people improve it. Farhadi also argues that open-source models can bring returns on AI investment where large, expensive proprietary models will not. These are the speakers’ claims, not established outcomes. Jernite says techniques for producing higher-quality training data could help govern data used in AI development. The report does not give the full test methods or results, so how Molmo’s performance compares beyond the described tests remains unknown.
Original source
MIT Technology Review · Article
Source details
Original source date: Sep 25, 2024 (2024-09-25T13:00:00Z).
Date basis: Date from the publisher’s own records, through its API.
Added to Escape Meridian: 2026-11-17T02:18:24.203Z. First observed: 2026-10-30T06:53:59.581Z. Retrieved: 2026-10-30T06:55:46.192Z.
Limits
Text as retrieved on October 30, 2026; originally published September 25, 2024. The report describes performance tests but does not give their full methods or results, limiting assessment of the comparison. The report’s described demonstration includes a failure to locate a specific parking lot; it does not establish broader real-world reliability.
Open question
How will applications built on Molmo, and the ways people improve it, shape its significance beyond the reported tests?
Source-stated
MIT Technology Review reports that Ai2 says Molmo performs as well as top proprietary models on tests involving images, charts and documents.
The Allen Institute for Artificial Intelligence (Ai2), a research nonprofit, is releasing a family of open-source multimodal language models, called Molmo, that it says perform as well as top proprietary models from OpenAI, Google, and Anthropic.
Source-stated
Ai2 attributes Molmo’s performance with fewer resources to a smaller, curated image dataset and detailed human descriptions converted from speech into training data.
Ai2’s Molmo models have been trained on a significantly smaller and more curated data set containing only 600,000 images
Source-stated
In a demonstration described by MIT Technology Review, Molmo counted deck chairs but could not locate a specific parking lot.
It was not perfect, however. It could not locate a specific parking lot, for example.
Source-stated
Ai2 CEO Ali Farhadi argues that open-source models can use money and time efficiently and can bring returns on AI investment where large, expensive proprietary models will not.
Big, expensive proprietary models won’t do that, he argues, but open-source ones can. He says the work shows that open-source AI can also be built in a way that makes efficient use of money and time.
Source-stated
Hugging Face’s Yacine Jernite says Molmo’s significance will depend on the applications developers build and the ways people improve it.
its real significance will lie in the applications developers build on top of it, and the ways people improve it.
Source and contributor context
MIT Technology Review · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Multimodal model
A model that works with more than one kind of input or output, such as text, images and audio.
Ai2
Ai2 develops AI research tools and models, including the openly documented Olmo family.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Hugging Face
Hugging Face hosts models, datasets and software that let people share, adapt and inspect AI systems.
Editorial provenance
Version 1. Publication ID: dca0a5f7-d96e-b864-364f-36382ab6b824.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-17T02:13:56.992Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: d03c408de497422f408f394c63ab260fdab90daf489b4af6ab1a3d4f344d9540. Source text hash: d27367694081826976861f539bdb7db6f93750bf49e81fb3251b07f1f4ce1c29.
passed · 2026-11-17T02:13:56.992Z · anthropic/claude-opus-5.5
Publication: dca0a5f7-d96e-b864-364f-36382ab6b824
attribution — pass: The reading attributes the report to MIT Technology Review and keeps the performance claims as Ai2's. It credits the views to Ali Farhadi, Yacine Jernite and Ai2 correctly. The evidence is a single article, not a listing.
that it says perform as well as top proprietary models from OpenAI, Google, and Anthropic
summary — pass: The reading leads with the actor (Ai2) and the development (open-source Molmo models). "Multimodal" is defined in the page definitions, and "point" is explained in plain language. It keeps the demo's failure, Farhadi's argument about investment returns and Jernite's caveat. The question follows from Jernite's stated view and is not leading. A reader without specialist background can follow it.
its real significance will lie in the applications developers build on top of it, and the ways people improve it.
date — pass: September 25, 2024 matches the UTC day of originalPublished. The page was modified in December 2024, but the reading relies only on the main article body. Later sidebar items, such as the AlphaGo ten-year piece, are not used. Forecasts and arguments are kept as attributed claims. Kind is publisher-page, so no archive disclosure is needed.
support — pass: Each claim matches the article body: the curated 600,000-image dataset, the spoken descriptions converted to data, the deck-chair counting and the parking-lot failure, Farhadi's argument about investment returns and Jernite's view on significance. The cited text is plainly part of the original article, so the later edit does not cast doubt on it. The stated limitations do not overreach.
Ai2 achieved this by getting human annotators to describe the images in the model’s training data set in excruciating detail over multiple pages of text.
media — pass: No media is selected, and no verified media candidates were supplied. Omitting media is appropriate.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
Ai2 develops AI research tools and models, including the openly documented Olmo family.
Explore Ai2A model that works with more than one kind of input or output, such as text, images and audio.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEHugging Face hosts models, datasets and software that let people share, adapt and inspect AI systems.
Explore Hugging FaceAi2 develops AI research tools and models, including the openly documented Olmo family.
Explore Ai2A model that works with more than one kind of input or output, such as text, images and audio.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEHugging Face hosts models, datasets and software that let people share, adapt and inspect AI systems.
Explore Hugging Face