GLM-4.5 Air glm-4.5-air
zai-org · glm hybrid-reasoning efficient
#24 overall
- params
- 106B (A12B)
- arch
- moe
- context
- 128k
- license
- mit open weights
- released
- Jul 2025
- reasoning
- yes
- downloads/30d
- 418.4k
# why ranked
Overall: #24 — score 28.7 ● medium — one signal missing, weights renormalized
| signal | weight | input (0–100) |
|---|---|---|
| aa_intelligence | 0.50 | 16.5 |
| arena_elo | 0.30 | — |
| bench_composite | 0.20 | 59.1 |
missing signals are dropped and the remaining weights renormalized — never imputed.
# trend
collecting history — trends appear after a few daily runs.
# benchmarks
| benchmark | score | source | date |
|---|---|---|---|
| AA Intelligence Index | 16.5 | Artificial Analysis | — |
| AA Math Index | 80.7 | Artificial Analysis | — |
| Aa-index | 0.6 / 1 | LLM Stats | — |
| AIME 2024 | 0.9 / 1 | LLM Stats | — |
| Bfcl-v3 | 0.8 / 1 | LLM Stats | — |
| Browsecomp | 0.2 / 1 | LLM Stats | — |
| Math-500 | 1.0 / 1 | LLM Stats | — |
# where to run
| provider | quant | ctx | $/M in | $/M out | $/M blend | tps | uptime |
|---|---|---|---|---|---|---|---|
| Novita | bf16 | 128k | $0.13 | $0.85 | $0.31 | — | 99.7% |
| SiliconFlow | fp8 | 128k | $0.14 | $0.86 | $0.32 | — | 99.5% |
| Z.AI | fp8 | 128k | $0.20 | $1.10 | $0.43 | — | 97.3% |
blended = (3·input + output) / 4 per 1M tokens · cheapest first
source aliases
- aa
glm-4.5-air- arena
GLM-4.5-Air- openrouter
z-ai/glm-4.5-air