Post #1615171
2026-04-22 16:15 UTC
Replies (4)
-
@rademaker@podcastindex.social 2026-04-22 16:41
@dave opus seems to be miles ahead of any open source local model. My dream is Meta gets Llama on par just in time for Apple to release M5 max chips in a Mac Studio (which should be powerful enough?)
-
@theDanielJLewis@podcastindex.social 2026-04-22 17:12
@dave Meh. GPT seems to actually be consistently better for coding anyway, and we have yet to see GPT 5.5. Plus, other models are getting surprisingly good, like the new Kimi K2.6. And I saw that Mimo v2.5 Pro was released just today. I never used that model for coding, but it _did_ do really well for OpenClaw agentic work before I switched to GPT via a ChatGPT $20/month subscription. Cursor, however, appears to cap my legacy subscription to GPT 5.3 Codex and Opus 4.6.
-
@mitch@podcastindex.social 2026-04-22 19:38
@dave I've been running my LLM accounts hard this whole year, in part because I suspect the prices will go up dramatically at some point. I mainly use Cursor (the max $200 per month plan). Their "Auto" model is great for most tasks, and the most efficient option apparently. For more complex / higher stakes tasks I have been using Codex 5.3. I don't know if it is better or worse than Anthropic models at this point, but I have no complaints about how it is working for me.
-
@martin@podcastindex.social 2026-04-23 10:59
@dave Claude Code has become almost unusable for me. My coworkers do not seem to have the same problem, but I am on the $200/mth plan and my usage goes down like ABSOLUTELY crazy. In the past I could work on like 7 parallel projects and keep Claude Code going for all of them without running out of weekly usage. Now with just 1 project I run out after 4 days.