Top token user vouches for Carbon: strong at xhigh for complex coding
olcan · x · 2026-11-18
Developer doodlestein endorsed the Carbon model as "really strong on xhigh effort," finding it highly complementary to Opus/Fable/Astra and particularly adept at applying his reality-check skill to complex software projects.
Boosting the credibility of the take, olcan notes that Jeff ranks #1 on TokScale's human token leaderboard with 4.6T tokens (1.5T in the last month) and cycles through 28 Claude Ultra, 25 GPT Pro, and 7 Google Ultra subscriptions.
More from coding & agent
- Proactive Agents That Schedule Their Own Work Hinge on Memory and UX, Says LangChain's Runkle — hwchase17 · 2026-11-18
- Cognition Launches Devin Bots: One Bot Leads a Team of Agents — aidangch · 2026-11-18
- 2026's 10 Best AI Agent Frameworks: Agent Adoption Doubled, 90% Now Write Data — rseroter · 2026-11-18
- Replit Livestream With BabyAGI Creator Yohei Nakajima on Building Frontier Agents — ColleenMBrady · 2026-11-18
- fx: a tiny open-source Zig coding agent with a 6.33 MiB binary that runs in the browser — evilrabbit_ · 2026-11-18
- Dev gripes: Anthropic models push temp folders over local-dir instructions — alexcovo_eth · 2026-11-18