maternion/ context-1:20b-q4_K_M

9 yesterday

Context-1 is a 20B agentic search model that breaks complex queries into steps, retrieves relevant info iteratively, and manages its context to support multi-hop reasoning alongside a larger model.

tools thinking 20b
ollama run maternion/context-1:20b-q4_K_M

Details

yesterday

ca13ad6c97ad · 16GB ·

gpt-oss
·
20.9B
·
Q4_K_M
{ "temperature": 1 }
<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI. Knowledge cutof

Readme

Chroma Context-1

Context-1 is a 20B parameter agentic search model trained to retrieve supporting documents for complex, multi-hop queries. It is designed to be used as a retrieval subagent alongside a frontier reasoning model: given a query, Context-1 decomposes it into subqueries, iteratively searches a corpus, and selectively edits its own context to free capacity for further exploration.

Context-1 achieves retrieval performance comparable to frontier LLMs at a fraction of the cost and up to 10x faster inference speed.

Technical report: Chroma Context-1: Training a Self-Editing Search Agent

Model Details

  • Base model: gpt-oss-20b
  • Parameters: 20B (Mixture of Experts)
  • Training: SFT + RL (CISPO) with a staged curriculum
  • Precision: BF16 (MXFP4 quantized checkpoint coming soon)

Key Capabilities

  • Query decomposition: Breaks complex multi-constraint questions into targeted subqueries.
  • Parallel tool calling: Averages 2.56 tool calls per turn, reducing total turns and end-to-end latency.
  • Self-editing context: Selectively prunes irrelevant documents mid-search to sustain retrieval quality over long horizons within a bounded context window (0.94 prune accuracy).
  • Cross-domain generalization: Trained on web, legal, and finance tasks; generalizes to held-out domains and public benchmarks (BrowseComp-Plus, SealQA, FRAMES, HLE).

Important: Agent Harness Required

Context-1 is trained to operate within a specific agent harness that manages tool execution, token budgets, context pruning, and deduplication. The harness is not yet public. Running the model without it will not reproduce the results reported in the technical report.

We plan to release the full agent harness and evaluation code soon. In the meantime, the technical report describes the harness design in detail.

Citation

@techreport{bashir2026context1,
  title = {Chroma Context-1: Training a Self-Editing Search Agent},
  author = {Bashir, Hammad and Hong, Kelly and Jiang, Patrick and Shi, Zhiyi},
  year = {2026},
  month = {March},
  institution = {Chroma},
  url = {https://trychroma.com/research/context-1},
}

License

Apache 2.0