Weave Router alternative

Before you switch routers, verify the savings.

Weave Router is a real, drop-in LLM router product with solid engineering behind it. OmnisBench isn't a competing router — it's the open, reproducible way to check whether routing claims like theirs hold up, and it's also where you'll hear about OmnisRouter, an open router coming soon from the same team.

Why people search for this

Why people look for a Weave Router alternative

A few things come up consistently when people evaluate Weave Router:

Closed, black-box routing

Requests are routed by an in-process cluster scorer with unpublished centroids. You can see that a decision was made, not why.

Unverified savings claims

Marketing cites 40–70% savings (and up to 80–85% in some copy), but there's no published benchmark methodology or dataset backing the number.

Elastic License v2, not open source

The code is source-available under ELv2, not an OSI-approved open source license — a meaningfully different trust posture than a permissive license.

Managed unified-credit billing

The managed tier bills through unified usage-based credits with no public price list, which reads as lock-in to teams used to transparent per-model pricing.

The honest framing

OmnisBench is not a router — it's how you verify one

OmnisBench doesn't sit in your request path and doesn't route traffic. It's an open, Apache-2.0 benchmark that scores routing policies against a quality-per-dollar frontier, publishes every underlying model response, and lets anyone re-grade the results offline. If a router — Weave Router or otherwise — claims a savings percentage, OmnisBench is the tool built to check it rather than take it on faith.

How to check a router's claims

Verifying routing savings with OmnisBench

Two ways to use it: run the benchmark yourself against the same task suite and model pool a router claims to route across, and compare the resulting frontier to the router's advertised savings; or, if a routing benchmark's results are already published, run omnisbench verify to independently re-grade them offline, with zero API calls and no keys.

# re-grade a published run yourself, no API keys needed
$ pip install -e .
$ python -m omnisbench.cli verify --run runs/2026-08-19
VERIFY OK

Category comparison

Two different kinds of product

Weave Router and OmnisBench aren't really competitors — they're different categories. Here's both, stated fairly:

Dimension	Weave Router router product	OmnisBench open benchmark
What it does	Routes your live traffic between models	Grades routing policies; doesn't route traffic
Routing logic	Closed, in-process cluster scorer (unpublished centroids)	N/A — publishes model responses for independent grading
Savings claims	40–70% claimed (marketing up to 80–85%), no published benchmark	Publishes results.json; every number re-gradable offline
License	Elastic License v2 (source-available)	Apache-2.0 (OSI open source)
Billing	Managed unified usage-based credits (self-host BYOK also free)	Free, self-run CLI
Enterprise posture	SOC2 Type II, VPC/air-gap self-host, funded team	Community project; not an enterprise product

To be fair to Weave Router: the engineering is genuinely solid — in-process ONNX routing, cross-format cache-warmth preservation, session pinning, and a frictionless drop-in endpoint swap. Those are real, non-trivial capabilities.

If you want an open router

OmnisRouter — coming soon

The same team behind OmnisBench is building OmnisRouter, an open router that applies the same verify-first philosophy to actual traffic routing. It isn't released yet, so there's nothing to link or install today — watch this site for it. In the meantime, if you need a router now, worthwhile options to evaluate include Weave Router itself (see who it fits, below) and OpenRouter's built-in auto-router.

Who Weave Router fits

Who Weave Router is a good fit for

Being fair to them: if you want a managed, SOC2-audited, drop-in router with VPC/air-gap self-hosting, a funded team behind it, and you're comfortable with closed routing logic and usage-based credit billing in exchange for that convenience, Weave Router is a reasonable choice — particularly for teams already inside the agentic-coding tool ecosystem it plugs into (Claude Code, Codex, Cursor, opencode).

See the numbers for yourself

Check OmnisBench's own first-run leaderboard, or pull the source and re-grade it offline.

See the leaderboard → · GitHub →

Questions

Frequently asked

Is OmnisBench a Weave Router alternative?

Not directly — OmnisBench is a benchmark, not a router, so it doesn't route your traffic the way Weave Router does. What it does do is give you an open, reproducible way to check routing-savings claims like Weave Router's. If you want an actual open router, OmnisRouter from the same team is coming soon, built on the same verify-first philosophy.

Why do people look for a Weave Router alternative?

Common reasons: Weave Router's routing model and centroids are closed-source so you can't inspect how decisions are made; its 40-70% (marketing up to 80-85%) savings claims have no published benchmark backing them; it ships under the Elastic License v2, which is source-available rather than OSI open source; and its managed tier uses unified usage-based credit billing, which can feel like lock-in.

How do I check whether a router's savings claims are real?

Run OmnisBench against the same task suite and model pool a router claims to route across, or use omnisbench verify to independently re-grade any published results. Because OmnisBench publishes the model response for every task, you can confirm the quality and cost numbers yourself instead of taking a vendor's percentage on faith.