davsclaus opened a new pull request, #1804: URL: https://github.com/apache/camel-website/pull/1804
The third post in the local-model benchmark series, after [the first](https://camel.apache.org/blog/2026/09/camel-local-model-benchmark/) and [round 2](https://camel.apache.org/blog/2026/09/camel-local-model-benchmark-round-2/). It leads with Kamelets: - A local model knows Camel's components from memory, but hardly knows Kamelets: 3 of 15 steps bare and 1 of 15 with tools before the fixes. - Once the catalog tools and the validator knew Kamelets (CAMEL-25379 and follow-ups), Kamelets passed 47 of 50 steps with tools, ahead of the components they wrap (37 of 50). - The model wrote its own Kamelets 19 times out of 20 with tools: custom Kamelets as building blocks for agents. Then the step-by-step ladder (21 examples, 71 steps, 5 runs each): 84.5% (r3) to 92.4% (r7, Oct 9), with the fixes behind it. It closes with the method: real tests and benchmarks, and many small fixes that help both people and agents, which we will keep applying to other parts of Camel. All numbers are from the runs of Oct 2-9 on Camel main; the fixes ship in Camel 4.23. The featured image is the one from round 2. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01STT6whBgK1AqsSsUKrnE8m -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
