Can it run here, and can it actually do anything?
OpenAI released gpt-oss on 5 August and GPT-5 on 7 August. On 9 August, my questions were about local Ollama models that could work with terminal tools, then about the new gpt-oss models.
Retrospective field note for 2025-08-11. Reconstructed in November 2026 from dated sources, not originally published on the archive date.
OpenAI released gpt-oss on 5 August and GPT-5 on 7 August. On 9 August, my questions were about local Ollama models that could work with terminal tools, then about the new gpt-oss models.
That’s a much more concrete question than which model is winning this week. Can it run on the machine available? Can it use the tools? Can it finish a job without creating another one?
The archived chat shows me exploring those options. It doesn’t show a completed deployment or a benchmark, and I’m not going to retrofit one into the story.
Local inference changes the choices around hosting and access. It doesn’t supply the rest of the agent. There still needs to be a tool interface, a permission boundary, and somewhere sensible to go when a command fails.
I’d compare candidates on one real task with the same files and the same checks. Count the repairs as well as the successful calls. If the fast model needs constant correction, its speed isn’t doing much for the person waiting.
Hardware claims need checking too. A laptop’s product name is not a specification sheet. Start with what is actually installed, then work out what fits. Less leaderboard. More working terminal.