Google: Gemma 3 27B
google/gemma-3-27b-it
Cheapest provider
$0.08 / 1M
DeepInfra
Fastest provider (p95)
63 tok/s
Phala
Intelligence (composite)
66.3
MMLU-Pro · HumanEval · math · GPQA
Per-provider performance
Latency / throughput / uptime measured across providers over the last 30 minutes of live traffic. Atlas’s router weighs these per call (with the eval-gate signal) when picking a variant for Standard and Batch tiers.
| Provider | Quant | Input $/1M | Output $/1M | Latency p50 / p95 | Throughput p50 / p95 | Uptime 30m | Success |
|---|---|---|---|---|---|---|---|
| DeepInfra | q8· fp8 | $0.0800 | $0.1600 | 908ms / 3972ms | 18 / 28 tok/s | 97.94% | 97.2% |
| Parasail | q8· fp8 | $0.0800 | $0.4500 | 984ms / 3131ms | 31 / 54 tok/s | 99.91% | 99.2% |
| Nebius | q8· fp8 | $0.1000 | $0.3000 | 643ms / 1851ms | 17 / 54 tok/s | 97.51% | 99.5% |
| Phala | undisclosed | $0.1100 | $0.4000 | 744ms / 1837ms | 34 / 63 tok/s | 100.00% | 61.1% |
| Novita | full· bf16 | $0.1190 | $0.2000 | 990ms / 38881ms | 18 / 29 tok/s | 71.89% | 90.2% |
“—” 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
64.3
broad reasoning
Code
79.2
pass@1 (HumanEval / LiveCodeBench)
MATH
75.2
math accuracy
GPQA Diamond
41.8
hard reasoning
Source: Gemma 3 technical report
How Atlas routes Google: Gemma 3 27B
- Realtime tier — direct passthrough at the upstream’s native precision. Best for hard latency / quality guarantees.
- Standard tier — Atlas picks the cheapest provider variant whose quantization has stayed green on your operation’s eval gates. For Google: Gemma 3 27B that’s currently DeepInfra at $0.08/1M.
- Batch tier — async, biggest discount. Roadmapped to use provider batch APIs (OpenAI / Anthropic) where available and queued spot capacity for open-weight workloads.