Databricks CTO Matei Zaharia says open-source GLM 5.2 on the minimalist Pi harness tied Claude Opus 4.8 at 35% lower cost
The evaluation suggests vendor harnesses inflate token consumption and operational costs.
Combined views
955.6K
24 Sources, first seen ago
Sources
ME 
merve@mervenoyannvery cool morning read > Pi has slightly higher pass rate with Opus 4.8 than Claude Code, and sends less context (likely does better compaction) > Pi with GLM 5.2 is better than GPT 5.5 and less costly than Opus + Pi > same pattern with Codex too, just Opus 4.8 is a better…
- likes: 672
- replies: 20
- bookmarks: 503
- reposts: 39
RP Rohan Paul@rohanpaul_aiSurprising and such a good news for open source coding model, and also that there are lots of hidden chances to reduce cost while improving quality. Databricks showed GLM-5.2 can compete with elite closed coding models inside real enterprise code. GLM 5.2 landed in…
- likes: 57
- replies: 12
- bookmarks: 16
- reposts: 10
AK Andy Konwinski@andykonwinskiagent architectures have clear parallels to monolithic operating systems vs micro-kernels. and we can get even more minimal agent "kernels" than pi https://twitter.com/matei_zaharia/status/2074943612631273730
- likes: 16
- replies: 4
- bookmarks: 4
- reposts: 0
DJ Daniel Jeffries@Dan_Jeffries1Open Source is the... https://twitter.com/matei_zaharia/status/2074943612631273730
- likes: 8
- replies: 0
- bookmarks: 2
- reposts: 1
EL elvis@omarsar0Highly-recommended read. Aligns with what I see in my own harness: > Pi harness got the same success rate as harnesses from the LLM vendors with Opus and GPT, but at 2x less cost > GLM 5.2 was a major step forward in open-source coding agent performance The harness bloat/rot…
- likes: 438
- replies: 27
- bookmarks: 429
- reposts: 49
TW Thomas Wolf@Thom_Wolf- likes: 57
- replies: 4
- bookmarks: 44
- reposts: 1