Each suggestion would name a candidate for one of the agents' roles (driver, helper, expert, classifier, reviewer). It would then go through a benchmark before any role changes, as qwen3.8-27b did in PR #55.
To settle:
Who checks the leaderboard, and how often. An agent might do it later, perhaps nisse-ops as a duty (O5).
What turns a candidate into a benchmark run: for example, it fits the 4090, LiteLLM can serve it, and it beats the current pick for a role.
Where suggestions land: an issue per candidate, or a comment here.
First candidate, already noted: Qwen3.8-Flash-Next, to be benchmarked against qwen3.8-27b.
The operator wants to start using the whatmodel.sh leaderboard as a source of suggestions for model updates:
- AI model leaderboard: the best AI models you can run · whatmodel.sh — https://share.google/vTQVizL9PVnQ3zHky
Each suggestion would name a candidate for one of the agents' roles (driver, helper, expert, classifier, reviewer). It would then go through a benchmark before any role changes, as `qwen3.8-27b` did in PR #55.
To settle:
- Who checks the leaderboard, and how often. An agent might do it later, perhaps nisse-ops as a duty (O5).
- What turns a candidate into a benchmark run: for example, it fits the 4090, LiteLLM can serve it, and it beats the current pick for a role.
- Where suggestions land: an issue per candidate, or a comment here.
First candidate, already noted: **Qwen3.8-Flash-Next**, to be benchmarked against `qwen3.8-27b`.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_01J5Utyfzw7gF7UhjRDEtjTk
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
The operator wants to start using the whatmodel.sh leaderboard as a source of suggestions for model updates:
Each suggestion would name a candidate for one of the agents' roles (driver, helper, expert, classifier, reviewer). It would then go through a benchmark before any role changes, as
qwen3.8-27bdid in PR #55.To settle:
First candidate, already noted: Qwen3.8-Flash-Next, to be benchmarked against
qwen3.8-27b.🤖 Generated with Claude Code
https://claude.ai/code/session_01J5Utyfzw7gF7UhjRDEtjTk