← Back to AI Insights
Gemini Executive Synthesis

Weave Router 2.0, an open-source model router for coding agents.

Technical Positioning
Outperforms single LLMs with an ensemble approach, achieving Astra-level performance at significantly lower cost (52-54% of Astra's cost) and faster execution (2.2-2.5x faster). Positions itself as a cost-effective and performance-optimized alternative/enhancement for existing coding agents.
SaaS Insight & Market Implications
This submission addresses critical pain points in LLM-driven development: cost and performance optimization. The Weave Router 2.0 demonstrates a viable strategy for achieving "frontier model" performance (e.g., Astra) at half the cost and over twice the speed by intelligently routing tasks across an ensemble of LLMs. This validates the multi-model, specialized approach over reliance on a single monolithic model. The technical innovations in architecture (HMM + classifier for session state) and cache management are key differentiators, directly impacting operational efficiency and expenditure. For B2B SaaS, this represents a significant opportunity to reduce infrastructure costs for AI-powered coding tools and accelerate development cycles, making advanced AI capabilities more economically accessible and performant. The open-source nature combined with a hosted version offers flexible adoption paths.
Proprietary Technical Taxonomy
model router coding agents ensemble approach LLMs Claude Code Codex GPT-6 Astra Terminal Bench 4.0

Raw Developer Origin & Technical Request

Source Icon Hacker News Oct 2, 2026
Show HN: Open-source model routing for coding agents at Astra-level performance

A few months ago we started building a model router for coding agents because we thought we could outperform any single model with an ensemble approach. Recently we’ve achieved that milestone and I want to talk about how we did it.First of all, a quick explanation: the Weave Router (github.com/weave-os/router plugs into any coding agent (e.g. Claude Code or Codex) and intelligently switches between LLMs. So, for example, Astra handles tricky debugging or complex system design tasks, and Deepseek v4 Flash handles simple frontend updates.What we’re announcing today is our new routing model, which we’re calling Weave Router 2.0. We benchmarked 2.0 against GPT-6 Astra on Terminal Bench 4.0 and SWE Atlas. On both benchmarks, the router had equivalent pass rates. On Terminal Bench, the router hit 52% of Astra’s cost, and completed tasks 2.2x faster. On SWE Atlas, the router cost 54% as much as Astra and ran 2.5x faster. (Full results on our website at weaveos.com/router!)It turns out training a model to route effectively - taking into consideration model capabilities, costs, cache awareness, and more - is a really hard problem! I want to talk about three ways we were able to improve so much over the last few months: 1) a new architecture, 2) larger training data set size, and 3) smarter cache-eviction impact calculation.1) a new architecture. Our initial approach used an RL model without many priors. While RL is still an important part of the story, the cost of fully exploring the space of routing decisions is very high, so we’ve taken some shortcuts that have significantly improved performance.Consider how large the search space for the routing problem is. Take a typical coding agent session, with ~100 agent turns (i.e. 100 LLM API calls). Technically there are 100 chances to select a model. If we assume a roster of ~10 models (of course there are lots more but we can remove any that are Pareto dominated), then there are 10^100 possible paths through that session. We simply cannot explore all of them! So that's why clever tricks to shrink this space are so important.In particular: we trained a hidden Markov model to trace the session state, then a classifier maps the session to one of a few buckets of similar models. Using the HMM allows us to evaluate not just where a session is currently, but how it got there. We've gotten significantly better performance on bucket selection by incorporating that information - we believe this is because two sessions that might look quite similar to a naive classifier are much better distinguished by this HMM approach.Using this HMM + classifier to select a bucket first significantly shrinks the space to explore, by throwing out most models that could not reasonably serve the given session. This rearchitecture was the single biggest performance unlock!2) larger training data set size (much less technically interesting but still an important part of the story). By using frontier LLMs to help us label a larger and more diverse set of coding agent sessions, we were able to bootstrap the two models discussed in 1) to a better state, while also providing even richer reward signals for RL.3) smarter cache-eviction impact calculation. One of the hardest parts of routing well (if you care about saving money) is using the model caches intelligently. We built a subsystem that can calculate the expected value of switching models (and thus paying a high one-time cost to fill up a different cache) much more accurately, helping us avoid costly and unnecessary switches in more cases, while still switching when the benefit outweighs the cost. This is where most of our improvement on cost has come from.We still have a lot of room to continue to improve (we won’t rest until we’re consistently beating Astra/Fable, not just tying!) but matching frontier model performance was a huge milestone for our routing model, and in my opinion validates our initial hypothesis that an ensemble of models can do better than any single model ever could.Our router is open source (github.com/weave-os/router so anyone can try it out. Or if you prefer you can use our hosted version (weaveos.com/router

Developer Debate & Comments

capocasa • Oct 2, 2026
That sounds really quite interesting- one question pops up, this sounds like it could produce a lot of non-cache input tokens. How much impact does this have, and are you mitigating it?Interested in adding support to my own 3code- https://3code.capocasa.dev- which aims to reduce cost by using more compact system prompts and a clever compaction variant.
devmor • Oct 2, 2026
I would like to see this (and any model router, frankly) benchmarked against two things, personally:1) Claude Code's "advisor mode" (nominally, Sonnet 5.5 with a Fable advisor)2) Copilot's "HydraFusion" model router/advisor combo.Specifically I would like to see them compared on architecture planning (both human assisted and hands-off with a draft document) and code review, as these are what I have found the most significant improvement on with multi-model systems.> our initial hypothesis that an ensemble of models can do better than any single model ever could.To your hypothesis, anecdotally I find both of these offerings to be far superior to any single model for most tasks of any real complexity, and both to have general frustration/failure cases that single models do not. I would not be surprised that any ensemble approach that utilizes more than one single model meets this hypothesis.Quite frequently I delegate review and restructuring loops to subagents acting as judges/advisors to tell the primary agent if it met the goal it was instructed to. For some workloads, I will even vet every tool call and user-facing output this way.
creativeSlumber • Oct 2, 2026
> then there are 10^100 possible paths through that session.Is this correct? in practice you would only be deciding what the next turn is going to be. Because the turn after the next turn is determined by the outcome of the next turn. So shouldn't this number be 10 * 100?
svnt • Oct 1, 2026
If you are training on data labeled by frontier models, how do you expect to exceed the performance of frontier models, other than in the cost dimension by recognizing simpler problems and routing to cheaper models?
gitowiec • Oct 1, 2026
Can it route to locally or LAN hosted Qwen or some other open weights model?
YuechenLi • Oct 1, 2026
Have you compared this to using GPT-6.1 Sol instead of GPT 6 Astra + Deepseek? From my test, 6.1 Sol is a lot more token efficient than 6 Sol while being similar to Astra in performance, and I don't really find 6 Astra to be significantly better than 6/6.1 Sol for general coding as I feel 6 Astra is only noticeably better at spatial reasoning/vision compared to 6 Sol, and 6.1 Sol really closed the gap on that front.
ajspig1 • Oct 1, 2026
How do you handle provider variance on OpenRouter for the opensource models? Or do you use your own hosted version to mitigate this?And for both opensource and closed source, does the router account for provider quality, or catch it when a provider degrades?
rirze • Oct 1, 2026
How does this choose which models to use with any arbitrary set of model providers to work from? And why is an openrouter necessary for self-hosting?
jamesforestwest • Oct 1, 2026
How do you define the model buckets, and what happens when a session genuinely needs a model that isn't in the bucket the HMM picked?
thefourthchime • Oct 1, 2026
Interesting work, and thanks for describing how your router works internally. It's definitely a fascinating subject. How would you say this compares to Cursor's auto mode?

Frequently Asked Questions

Market intelligence mapped to Weave Router 2.0, an open-source model router for coding agents..

What is the technical positioning of Weave Router 2.0, an open-source model router for coding agents.?
Based on our AI analysis of the original developer request, its primary technical positioning is: Outperforms single LLMs with an ensemble approach, achieving Astra-level performance at significantly lower cost (52-54% of Astra's cost) and faster execution (2.2-2.5x faster). Positions itself as a cost-effective and performance-optimized alternative/enhancement for existing coding agents.
How is the developer community reacting to Weave Router 2.0, an open-source model router for coding agents.?
Yes, we have tracked 42 direct responses and active debates regarding this specific topic originating from Hacker News.
Which technical concepts are associated with Weave Router 2.0, an open-source model router for coding agents.?
Our proprietary extraction maps Weave Router 2.0, an open-source model router for coding agents. to adjacent architectural concepts including model router, coding agents, ensemble approach, LLMs.

Engagement Signals

123
Upvotes
42
Comments

Cross-Market Term Frequency

Counts stored text matches for foundational terms like Claude Code and Codex in analysed public discussions. These counts do not measure adoption.

Stored text matches, not adoption or market-share measurements.

  • Claude Code: 159 records
  • Codex: 103 records
  • LLMs: 61 records
  • coding agents: 47 records
  • classifier: 8 records
  • model router: 3 records
  • GPT-6 Astra: 3 records
  • priors: 3 records
  • LLM API calls: 3 records
  • RL model: 2 records
  • session state: 2 records
  • frontier LLMs: 2 records
  • ensemble approach: 1 records
  • Terminal Bench 4.0: 1 records
  • SWE Atlas: 1 records
  • pass rates: 1 records
  • search space: 1 records
  • agent turns: 1 records
  • Pareto dominated: 1 records
  • hidden Markov model (HMM): 1 records
  • bucket selection: 1 records
  • training data set: 1 records
  • reward signals: 1 records
  • cache-eviction impact calculation: 1 records
  • model caches: 1 records
  • one-time cost: 1 records
  • frontier model performance: 1 records