Z.ai: GLM 4.6
z-ai/glm-4.6
Cheapest provider
$0.43 / 1M
DeepInfra
Fastest provider (p95)
69 tok/s
DeepInfra
Intelligence (composite)
85.4
MMLU-Pro · HumanEval · math · GPQA
Per-provider performance
Latency / throughput / uptime / price measured across providers over the last 30 minutes of live traffic. This is what proves “sourced cheapest” — Atlas mode draws on these per call to serve the cheapest path that holds quality.
| Provider | Quant | Input $/1M | Output $/1M | Latency p50 / p95 | Throughput p50 / p95 | Uptime 30m | Success |
|---|---|---|---|---|---|---|---|
| DeepInfra | q4· fp4 | $0.4300 | $1.7400 | 530ms / 1882ms | 34 / 69 tok/s | 100.00% | 100.0% |
| Novita | full· bf16 | $0.5500 | $2.2000 | 2227ms / 13329ms | 28 / 43 tok/s | 94.78% | 94.8% |
| AtlasCloud | q8· fp8 | $0.6000 | $2.2000 | 1680ms / 10957ms | 15 / 31.8 tok/s | — | 100.0% |
| Z.AI | undisclosed | $0.6000 | $2.2000 | 7387ms / 22333ms | 26 / 37 tok/s | — | 81.3% |
| Venice | q4· fp4 | $0.8500 | $2.7500 | 1159ms / 2000ms | 25 / 39.6 tok/s | — | 71.0% |
“—” means live telemetry hasn’t accumulated enough recent traffic for that endpoint. “undisclosed” means the provider serves the model but doesn’t expose the quantization label (typically running fp8 / int8 internally).
Intelligence breakdown
Composite score is a weighted average of public benchmarks (30% MMLU-Pro, 25% code pass@1, 25% math, 20% GPQA). Numbers come from model cards and the Artificial Analysis intelligence harness; missing components are renormalised over what’s present.
MMLU-Pro
84.6
broad reasoning
Code
82.8
pass@1 (HumanEval / LiveCodeBench)
AIME 2025
93.9
math accuracy
GPQA Diamond
79.1
hard reasoning
Source: Z.ai GLM-4.6 evaluations (MMLU-Pro 84.6, GPQA 79.1, LiveCodeBench v6 82.8, AIME25 93.9); docs.z.ai + github.com/zai-org/glm-simple-evals
How Atlas mode sources Z.ai: GLM 4.6
- Strict mode — pin Z.ai: GLM 4.6 exactly and we pass it straight through, sourced from the cheapest provider above. The same model, no substitutions — currently DeepInfra at $0.43/1M.
- Atlas mode — the default. Each call is auto-optimized for the cheapest path that holds quality, at least 5% off going direct from call one and climbing as it ramps. You always see which model served the call and exactly what you saved — thumbs-down anything you don’t like for a full refund.