Context-1
Agentic search at the Pareto frontier
Context-1, an open-weights model from Chroma, is trained with SFT and RL, and is competitive with frontier models at search, while being 10x faster and 25x cheaper.
Context-1
Inference speed is the bottleneck
Agents spend a large percentage of their time gathering information.
40 tok/s
278.5s
Opus 4.5
400 tok/s
28.8s
Context-1
0s
60s
120s
180s
240s
300s
LLM inference
search()
*illustrative example
