Perplexity and AWS both open‑source small decision models in one day
Perplexity released its OneJev decision models under Apache 2.0, and Amazon's Strands Labs shipped a 2-billion-parameter rival, both claiming speed and low price.
Perplexity chief executive Aravind Srinivas said on X the company is open-sourcing its multimodal Decision model and offering it through a Decisions API. He claimed it scores higher than Jev at a fraction of the price, at 4 cents per million input tokens. The post does not say which benchmark he means.
https://x.com/AravSrinivas/status/2105726953554927677
The OpenDecisions repository on GitHub describes open models and one server for the Decisions API. Its default model, OneJev, answers every question in a request in one forward pass and generates no text. The README lists sizes from 0.8 billion to 27 billion parameters. Code and weights are under Apache 2.0.
The README says OneJev-4B answers ten questions about one screenshot in 104 milliseconds on a single H200 GPU. That is the maintainers' figure. When we opened the repository it showed no stars, no forks and one commit, so there is no outside testing or community traction to report yet.
Separately, VentureBeat reported that Strands Labs, an experimental AWS agent project, released Strands Decider 2B on Thursday. It is fine-tuned from Alibaba's Qwen3.5-2B and is free on Hugging Face under Apache 2.0. VentureBeat says it ran at a median 106 milliseconds on an Nvidia RTX 3090.
VentureBeat reports accuracy of about 72 percent on the public JevBench and says the model ranks second among public models its size. Those are Strands Labs claims. They arrive after Cloudflare and Ollama released similar models this week, which suggests small decision models are becoming a crowded category.