Meta: Llama 3.1 70B Instruct
meta-llama/llama-3.1-70b-instruct
Cheapest provider
$0.40 / 1M
DeepInfra
Fastest provider (p95)
94 tok/s
WandB
Intelligence (composite)
66.5
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 | q8· fp8 | $0.4000 | $0.4000 | 241ms / 1283ms | 21 / 42 tok/s | 99.36% | 98.4% |
| DeepInfra | full· bf16 | $0.4000 | $0.4000 | 264ms / 998ms | 16 / 35 tok/s | 99.72% | 99.7% |
| Amazon Bedrock | undisclosed | $0.7200 | $0.7200 | 451ms / 775ms | 11 / 22 tok/s | 100.00% | 99.7% |
| WandB | full· bf16 | $0.8000 | $0.8000 | 310ms / 416ms | 54 / 94 tok/s | 100.00% | 100.0% |
“—” 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
66.5
broad reasoning
Code
80.8
pass@1 (HumanEval / LiveCodeBench)
MATH
68.1
math accuracy
GPQA Diamond
46.4
hard reasoning
Source: Meta Llama 3.1 model card (HumanEval / MATH / MMLU-Pro)
How Atlas mode sources Meta: Llama 3.1 70B Instruct
- Strict mode — pin Meta: Llama 3.1 70B Instruct exactly and we pass it straight through, sourced from the cheapest provider above. The same model, no substitutions — currently DeepInfra at $0.40/1M.
- 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.