Aria
Open multimodal mixture-of-experts model with under 4B active parameters
About
Aria is Rhymes AI's open multimodal mixture-of-experts model from October 2024, released under Apache 2.0. With about 25B total parameters and under 4B active per token, it handles text, images, video, and code in one model with a 64K context window. The team reports it beats Pixtral 12B and Llama 3.2-11B on vision benchmarks like MMMU and MathVista; it's a 2024 release, so newer open VLMs have since caught up on several tasks.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Team-reported: beats Pixtral 12B and Llama-3.2-11B on MMMU and MathVista; competitive with GPT-4o on long-video and document tasks.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Released October 2024 by Rhymes AI under Apache 2.0 (paper arXiv:2410.05993); ~25B total params, ~3.9B active per token; 64K multimodal context window; Benchmark comparisons (Pixtral 12B, Llama-3.2-11B, GPT-4o) are team-reported and labeled as such; 2024 release — entry notes newer open VLMs have caught up on several tasks; releasedAt null (day unknown). License: Apache-2.0 confirmed across sources..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Released October 2024 by Rhymes AI under Apache 2.0 (paper arXiv:2410.05993)
- · ~25B total params, ~3.9B active per token; 64K multimodal context window
- · Benchmark comparisons (Pixtral 12B, Llama-3.2-11B, GPT-4o) are team-reported and labeled as such
- · 2024 release — entry notes newer open VLMs have caught up on several tasks; releasedAt null (day unknown)