Skip navigation
Introducing Chroma Context-1, a 20B parameter search agent.
0:00 / 13:20

Introducing Chroma Context-1, a 20B parameter search agent.

Chroma

@trychroma31K followers

Visit on X
4,134 likesDownloadOriginal post

1,127,060 views

Introducing Chroma Context-1, a 20B parameter search agent. > pushes the pareto frontier of agentic search > order of magnitude faster > order of magnitude cheaper > Apache 2.0, open-source

40 Comments

Comments from the original post on X (Twitter)

  • Chroma's profile picture

    @trychroma

    Highly accurate search requires more than a single stage, the output of one search often informs the next. Frontier LLMs can approach these tasks through agentic search but long agentic search trajectories become cost and latency prohibitive. Context-1 solves these problems, delivering the best accuracy, speed, and cost across our generated benchmarks as well as public benchmarks like Browsecomp-Plus, SealQA, LongSealQA, and FRAMES.

    165 likes

  • Chroma's profile picture

    @trychroma

    Context-1 is trained on agentic search tasks using RL starting from gpt-oss-20b. Given a user query, it decomposes it into subqueries, searches iteratively, and returns a ranked set of supporting documents to a downstream reasoning model. It separates search from generation, each model performs the task it's best suited for.

    128 likes

  • Chroma's profile picture

    @trychroma

    One key technique the models harness leverages is self-editing context. As the agent searches, its context window fills with documents, many of which are irrelevant. Context-1 is trained to selectively prune its own context mid-search, freeing space for further exploration. As a result of being trained with this harness, Context-1 is far better than its base model at managing its own context window. This enables a 20B model with a 32k budget to outperform frontier models with larger context windows. The conservative token budget limits KV-cache invalidation costs, making mid-search pruning practical at serving time.

    182 likes

  • Chroma's profile picture

    @trychroma

    We trained Context-1 with SFT + RL using @thinkymachines Tinker on 8,000+ synthetic tasks across 4 domains: web, SEC filings, patent law, and email. Each task is multi-hop, requiring the agent to chain clues across documents. An extraction-based verification pipeline ensures label quality with high human-judge alignment. We're open-sourcing the full task generation codebase and weights of this model.

    200 likes

  • Chroma's profile picture

    @trychroma

    Context-1 matches or exceeds frontier models on retrieval performance. Results across our generated benchmarks and public evals (BrowseComp-Plus, SealQA, FRAMES, HotpotQA, HLE) demonstrate best in class performance. Search sub-agents are embarrassingly parallel. Running 4 parallel rollouts with rank fusion (Context-1 4x) for higher task performance is still cheaper than a single GPT-5 run.

    92 likes

  • Chroma's profile picture

    @trychroma

    We are releasing Context-1 as an open weights model, along with the full data generation pipeline used to train it. See our full report for a breakdown of our synthetic task generation, harness design, training methodology, along with an exhaustive evaluation of Context-1.

    107 likes

  • Max Rumpf's profile picture

    @maxrumpf

    We published our research in December and told Chroma's CEO Jeff. 4 months later, Chroma republished it without citing it. We think this sets a pretty bad precedent:

    64 likes

  • Jackie.W's profile picture

    @JackieWay2O

    This is the pattern I keep waiting for more teams to adopt — specialized smaller models as sub-agents instead of throwing frontier LLMs at every subtask. A 20B model that beats GPT-4.5 on search at a fraction of the cost is exactly what agent orchestration needs. The primary agent handles reasoning; the sub-agent handles retrieval. Each layer does what it's best at. The self-editing context trick (pruning irrelevant docs mid-search) is clever too. Context management is the real bottleneck in multi-hop agent tasks, not model capability.

    50 likes

  • 🔥 fire's profile picture

    @fire

    weights are here: ( link is inside the report, which you should read: )

    43 likes

  • Vaibhav (VB) Srivastav's profile picture

    @reach_vb

    gpt-oss 20b ftw! 🔥

    31 likes

  • Modal's profile picture

    @modal

    Incredible! 💚

    22 likes

  • Nick Khami's profile picture

    @skeptrune

    let's gooooo! cannot express enough how excited i am for this technology in general

    17 likes

  • Aamir's profile picture

    @aaxsh18

    this is super interesting. any chance you guys will release the code for the harness for the benchmarks?

    15 likes

  • Alessio Fanelli's profile picture

    @FanaHOVA

    I like the additional camera cuts to show that @kellyhongsn is not actually an AI model

    14 likes

  • dex's profile picture

    @dexhorthy

    very cool can I use this with claude code?

    13 likes

  • boris's profile picture

    @boristane

    what’s easiest way to try this model ? super interesting

    12 likes

  • Alex Volkov's profile picture

    @altryne

    Wohoo! Congrats! I wish you guys would release this a bit earlier to include on @thursdai_pod , but maybe next week @jeffreyhuber !

    10 likes

  • Eoghan McCabe's profile picture

    @eoghan

    That's awesome. Congrats

    7 likes

  • Henry Ventura's profile picture

    @hvent90

    SO COOL! > agent edits its own context super pumped to see how you folks did that. I'm experimenting with that right now. it's been on the back of my mind for months

    5 likes

  • Sriraam's profile picture

    @27upon2

    @willcb Omg this is so exciting and crazy that it’s open 🔥

    5 likes