Skip to content
Model Pareto

GLM-5.3-MXFP4

Red Hat AIOpen weightsMXFP4quantized from GLM-5.3huggingface.co ↗
Compare with nearest rivalsCompare with…
Input
—
Output
—
Speed
~78tok/sest.
Context
1Mtokens

Red Hat AI MXFP4 (W4A4, group 32, E8M0 scales) quant of GLM-5.3; ~465 GB on disk. bits 4.25 = 4-bit e2m1 + 8-bit shared scale per 32 weights, matching the other MXFP4 entries. Context from config.json max_position_embeddings 1048576 (card repo and base).

Rankings

Self-hostingEstimated

How we estimate →

What it would cost to run these open weights yourself on rented GPUs. No API sells this model, so this is its price on the chart.

Hardware
8× H200 141GB
MXFP4 weights
Throughput
~16,630 tok/s
many requests batched
Price
$0.25–2.03 /M tok
busy → light use
On your own machine
Mac Studio M5 Ultra 512GB (MXFP4, ~40 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • MXFP4 weights (744B params, 40B active per token) + 40% KV-cache headroom ≈ 553 GB
  • 8× H200 141GB at $3.59–$7.91/GPU-hour on-demand (2026-09-24)
  • ~16,630 output tok/s aggregate at batch 598 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~93,508 tok/s
  • Low = 75% utilization at the low GPU price; high = 20% at the high price

Benchmark scores

~ italic, dashed = no published score yet, estimated from related benchmarks.

Hard reasoning

Humanity's Last ExamEstimated
~#54 of 185 (estimated)~27.1%
  1. #1Claude Opus 5.561.4%
  2. #2Claude Fable 5.159.1%
  3. #53MiMo-V2.527.2%
  4. ~#54GLM-5.3-MXFP4Estimated27.1%
  5. #54Solar Mini 425.8%
No published score — estimated from related benchmarks
AA Intelligence IndexEstimated
~#32 of 86 (estimated)~28.0
  1. #1Claude Opus 5.557.6
  2. #2Claude Sonnet 5.556.0
  3. #31MiniMax-M329.2
  4. ~#32GLM-5.3-MXFP4Estimated28.0
  5. #32Quasar 438B26.7
No published score — estimated from related benchmarks