Agents on Rails: Maximum effort and DeepSeek 4.1 Flash

In the Stage 2 report we shipped 20 feature tickets on Fizzy and promised to explore benchmarking the agents all on max-effort. Now we have run it: every model on the board, same tickets, reasoning turned all the way up. We also tested a new model: DeepSeek 4.1 Flash . The short version: Effort matters, but only with some agents. GPT-6 Astra (already in first place) went from 35% to 53% success, GPT-5.6 Sol improved from 18% to 28%, and GPT-5.6 Luna performed much better: from 0% to 27% succe...