GLM-4.6 glm-4.6

zai-org · glm hybrid-reasoning flagship agentic

unranked — no quality signals yet

params
355B (A32B)
arch
moe
context
198k
license
mit open weights
released
Sep 2025
reasoning
yes
train compute
4.4e+24 FLOP

# why ranked

Coding: #7 — score 84.6 ● low — a single signal; treat with caution

signalweightinput (0–100)
aa_coding 0.25
aider_polyglot 0.25
arena_elo_code 0.15
swe_bench_verified 0.35 84.6

missing signals are dropped and the remaining weights renormalized — never imputed.

Agentic: #3 — score 84.6 ● low — a single signal; treat with caution

signalweightinput (0–100)
swe_bench_verified 1.00 84.6

full methodology

# trend

collecting history — trends appear after a few daily runs.

# benchmarks

benchmarkscoresourcedate
SWE-bench Verified 68.2 SWE-bench Sep 2025

# where to run

providerquantctx$/M in$/M out$/M blendtpsuptime
DeepInfrafp4 198k $0.43$1.74$0.76 99.9%
Venicefp4 198k $0.43$1.75$0.76 99.7%
Novitabf16 200k $0.55$2.20$0.96 96.0%
AtlasCloudfp8 198k $0.60$2.20$1.00 100.0%
Z.AIfp4 198k $0.60$2.20$1.00 56.3%

blended = (3·input + output) / 4 per 1M tokens · cheapest first

source aliases
aa
glm-4.6
arena
GLM-4.6
openrouter
z-ai/glm-4.6