Weave Router 2.0 claims GPT-6 Astra pass rates at about half the cost by switching models

Company benchmark claims; the HN thread tests them.

Weave’s open-source router switches a coding agent between models per task; the company reports pass rates equal to GPT-6 Astra on Terminal-Bench 4.0 and SWE Atlas at 52–54% of the cost.

Read it at the source

Hacker News (Show HN post)Show HN: Open-source model routing for coding agents ↗

Cascadic Analysis 2

UndertowMulti-agent systems can create cascading failures

What hidden risk could pull against this, even if the news is good?

A router hands each step to a different model, and each step assumes the previous output was sound. A mistake made by a cheaper model on an “easy” step can carry into the work a stronger model then builds on.

UndertowBenchmark scores can overstate real-world reliability

What hidden risk could pull against this, even if the news is good?

Equal pass rates on two benchmarks at about half the cost are the company’s own measurements. A router is tuned to the kinds of tasks it was trained and tested on; your tasks may not route the same way.

Weave Router 2.0 claims GPT-6 Astra pass rates at about half the cost by switching models

Further down the stream