New EXLR8 quant claims to beat EXL3 on every metric, GLM 5.3 cold-start TTFT at 30s
EAccelerate_42 · x · 2026-10-09
Developer mmastrac reports that EXLR8 (E-Accelerate), a new quantization format for GLM 5.3 full, is expected to beat EXL3 on every metric: better quality per size, faster speed per size, and quicker boot. The GLM 5.3 full server currently cold-starts to first token in 30 seconds, with further improvements expected.
More from Infra
- Hyperscaler signs 210MW 15-year lease, locking in roughly $5.2B in contracted revenue — YvesMulkers · 2026-10-09
- Analyst Doubles Down: NVL72 AI Servers Are the Biggest Product Cycle in History — firstadopter · 2026-10-09
- Pat Gelsinger: Memory, packaging and power—not chip speed—are AI's real bottlenecks — a16z Podcast · 2026-10-09
- Founder boosts CPU utilization with adaptive capacity planning instead of chasing users — DanielLockyer · 2026-10-09
- Open-source sfFFT delivers 3.8-6.1x speedup for FFT convolutions on DGX Spark GB10 — IgorCarron · 2026-10-09
- GPUs already within 2x of brain efficiency, and still beat human workers on energy — MikePFrank · 2026-10-09