Molmo-7B-D
Ai2's open vision-language model with pixel-level pointing, near GPT-4o on academic benchmarks.
About
Molmo-7B-D is the dense 7B variant of Ai2's Molmo family, released September 25, 2024. It pairs a vision encoder with a language model and a specialized pointing mechanism that lets it ground objects in images with pixel-level precision — useful for web agents, visual grounding, and reading analog displays like clocks and gauges.
Ai2 reports 77.3% average across 11 academic benchmarks (AI2D, ChartQA, DocVQA, TextVQA, MMMU, MathVista, and others), within striking distance of GPT-4o's 78.5, all under Apache 2.0. The model card notes the training data carries academic and non-commercial restrictions in places, so commercial users should check the dataset terms even though the weights are permissive.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- 77.3 average over 11 academic benchmarks (Ai2-reported)
- Human preference Elo 1056, vs GPT-4o 1079 (Ai2-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Ai2 release with strong community uptake (1600+ fine-tunes of the 8B successor); Apache 2.0 weights; training data has some academic/non-commercial restrictions noted on the model card; Ai2-reported 77.3 avg over 11 academic benchmarks vs GPT-4o 78.5. License: Apache 2.0 (weights); some training datasets carry academic/non-commercial terms.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Ai2 release with strong community uptake (1600+ fine-tunes of the 8B successor)
- · Apache 2.0 weights; training data has some academic/non-commercial restrictions noted on the model card
- · Ai2-reported 77.3 avg over 11 academic benchmarks vs GPT-4o 78.5