Skip to content
fedi.software

GLM 4.5 Air in fedi-index

The score in every slice, what it is made of, value, local run and history.

fedi-index · General

C 1,356 ±2

Percentile 49 · #151 · provisional

People prefer / Solves tasks

1,365 / —

Human votes vs benchmarks, Elo-eq

Value

-2

$0.425 per 1M tokens, median of 13 providers · vendor $0.425

On your hardware

  • GPU 12 GB #21 · Q4_0 · ≈5 tok/s
  • GPU 24 GB #21 · Q4_K_M · ≈6 tok/s
  • Mac 64 GB #21 · UD-IQ2_M · strong compression · ≈23 tok/s
  • 128 GB unified memory #28 · Q4_K_M · ≈11 tok/s

Score decomposition

C General 1,356 [1,353–1,358] Percentile 49 · #151 · components: 6 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/overall 1,384 ±2.16 1,384 ±2 26.1% 2026-10-02 —
LMArena text/hard_prompts 1,379 ±2.80 1,376 ±2 25.9% 2026-10-02 —
LMArena text/instruction_following 1,353 ±3.50 1,369 ±3 12.7% 2026-10-02 —
LMArena text/expert 1,369 ±7.93 1,370 ±6 11.2% 2026-10-02 —
LMArena text/longer_query 1,366 ±3.82 1,369 ±4 6.3% 2026-10-02 —
LMArena text/multi_turn 1,371 ±4.35 1,373 ±4 6.2% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

C Coding 1,361 [1,355–1,367] Percentile 48 · #145 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/coding 1,397 ±3.94 1,378 ±4 90.1% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Math 1,364 [1,352–1,377] Percentile 51 · #126 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/math 1,396 ±7.60 1,389 ±7 86.6% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Writing 1,359 [1,352–1,365] Percentile 51 · #133 · components: 2 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/creative_writing 1,343 ±4.85 1,370 ±5 61.4% 2026-10-02 —
LMArena text/longer_query 1,366 ±3.82 1,369 ±4 32.2% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

C Language: 繁體中文 1,383 [1,373–1,392] Percentile 50 · #113 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/chinese 1,427 ±6.48 1,395 ±5 93.5% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

C Language: Deutsch 1,367 [1,350–1,385] Percentile 38 · #89 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/german 1,376 ±11 1,382 ±10 91.9% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

B Language: English 1,374 [1,368–1,380] Percentile 55 · #124 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/english 1,397 ±2.91 1,385 ±3 93.9% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

C Language: Español 1,370 [1,352–1,388] Percentile 29 · #106 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/spanish 1,385 ±11 1,385 ±10 91.8% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

C Language: Français 1,368 [1,345–1,390] Percentile 28 · #97 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/french 1,398 ±14 1,385 ±13 90.3% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

C Language: 日本語 1,374 [1,360–1,389] Percentile 40 · #77 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/japanese 1,348 ±11 1,388 ±8 92.6% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

C Language: 한국어 1,352 [1,335–1,368] Percentile 30 · #96 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/korean 1,306 ±12 1,364 ±9 92.2% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

C Language: Polski 1,368 [1,356–1,379] Percentile 25 · #98 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/polish 1,367 ±7.23 1,380 ±6 93.2% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

C Language: Русский 1,371 [1,360–1,381] Percentile 46 · #127 · components: 1 provisional
Source Component Value Elo-eq Weight Measured Variant
LMArena text/russian 1,373 ±6.46 1,383 ±6 93.3% 2026-10-02 —

Weight is the share of the component in the score (the rest is the prior that pulls models with little data towards the population median). Methodology

History

fedi-index by month
The same in a table
Model 2026-10
GLM 4.5 Air 1,356 C

Badge

Show the model's fedi-index on your site or in a README. The badge updates itself; it links to this page.

fedi-index: GLM 4.5 Air

Data CC BY 4.0: keep the link to fedi.software.