GLM 5.2 — price by precision tier fp4 vs fp8 vs unknown

OpenRouter lists GLM 5.2 across 28 providers at 3 quantization tiers · cheapest provider per tier, per day · 2026-06-26 → 2026-07-05

fp8 (Jul3→Jul5)

-25.7%

fp4 (cheapest tier)

$1.96/M

fp8 now vs fp4

fp8 < fp4 (inverted!)

tiers tracked

① GLM 5.2 price by precision over time — cheapest provider per tier

The Jul 4–5 cut hit fp8 only; fp4 (the value tier) never moved. After the cut, higher-precision fp8 dropped below fp4 — a precision/price inversion.

② Precision spread today (2026-07-05) — price range within each tier

Min→max blended $/M across providers, with provider count. The cheapest fp8 provider now matches the canonical Z.ai price.

③ Frontier context — biggest price movers (2026-06-30 → 2026-07-05)

For reference: across the whole frontier, GLM 5.2 is the only model that meaningfully repriced. Outliers (|Δ|>60%, coverage artifacts) listed in footnote.

④ Frontier table (2026-07-05, sorted by Elo)

#	Model	Elo	In $/M	Out $/M	Blended $/M

Precision (fp4/fp8) from endpoints_panel.csv quantization × 26 providers; frontier/Elo from unified_intelligence_price.csv. Blended = (input+output)/2.