Claude Haiku 5.5 tops highlighted benchmark scores, likely far smaller than GLM 5.3 Flash
rickasaurus · x · 2026-10-08
scaling01 highlights the actual top benchmark scores and points out that Claude Haiku 5.5 is probably much smaller than GLM 5.3 Flash, implying better parameter efficiency. The comparison stems from a leaderboard screenshot by TheZachMueller, who noted the numbers were fetched by Codex and could be off.
More from Models
- Theo: viral $132M/year token cost claim is wrong — closer to $3M now, $1.2k soon — dotey · 2026-10-08
- X users call out xAI's pooled usage meters: video gen and inference shouldn't share one quota — TinfoilTricorn · 2026-10-08
- User claims Qwen3.8-Flash-Next-NVFP4 sounds uncannily like Opus 4.5 — natesiggard · 2026-10-08
- Dev endorsement: DeepSeek is the best bang-for-buck model, DSH an excellent harness — sull · 2026-10-08
- Creator's personal-assistant agent comparison: grok bot clearly best, Muse decent, Dots needs work — elonmusk · 2026-10-08
- Study Finds Embedding Models Often Ignore Retrieval Instructions; Distractor Fine-tuning Helps — _reachsumit · 2026-10-08