NEWMEN
← back to all models

Meta: Llama 3.3 70B Instruct

meta-llama/llama-3.3-70b-instruct

Metaopen-weight131K context15 providersIntelligence 58.4

Cheapest provider

$0.10 / 1M

DeepInfra

Fastest provider (p95)

521 tok/s

Groq

Intelligence (composite)

58.4

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.

ProviderQuantInput $/1MOutput $/1MLatency p50 / p95Throughput p50 / p95Uptime 30mSuccess
DeepInfraq8· fp8$0.1000$0.3200409ms / 3435ms15 / 29 tok/s99.06%99.8%
Inceptronq8· fp8$0.1200$0.3800586ms / 1914ms26 / 44 tok/s99.05%97.4%
AkashMLq8· fp8$0.1300$0.4000537ms / 1778ms31.5 / 65 tok/s99.42%99.9%
Nebiusq8· fp8$0.1300$0.4000409ms / 3175ms20 / 44 tok/s92.95%97.9%
Novitafull· bf16$0.1350$0.4000539ms / 1073ms34 / 46 tok/s99.97%100.0%
Parasailq8· int8$0.2200$0.5000698ms / 1322ms22 / 40 tok/s99.96%80.6%
Cloudflareq8· fp8$0.2930$2.2530569ms / 1952ms18 / 39 tok/s97.40%94.5%
SambaNovafull· bf16$0.4500$0.9000808ms / 3107ms34 / 203.8 tok/s99.54%97.4%
Groqundisclosed$0.5900$0.7900295ms / 776ms236 / 521 tok/s99.61%96.7%
Friendliundisclosed$0.6000$0.60000.0%
SambaNovafull· bf16$0.6000$1.2000808ms / 3107ms34 / 203.8 tok/s98.93%97.4%
WandBfull· fp16$0.7100$0.7100370ms / 1123ms37 / 90 tok/s100.00%100.0%
Googleundisclosed$0.7200$0.7200300ms / 1878ms18 / 62.6 tok/s88.21%83.8%
Googleundisclosed$0.7200$0.7200300ms / 1878ms18 / 62.6 tok/s97.79%83.8%
Togetherq8· fp8$0.8800$0.88001092ms / 2553ms32 / 78.6 tok/s97.56%97.1%

“—” 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

68.9

broad reasoning

Code

33.3

pass@1 (HumanEval / LiveCodeBench)

MATH

77.0

math accuracy

GPQA Diamond

50.5

hard reasoning

Source: Meta Llama 3.x cards — Llama 3.3 70B (MMLU-Pro 68.9, GPQA-Diamond 50.5, LiveCodeBench 33.3); llama.com

How Atlas routes Meta: Llama 3.3 70B Instruct

  • 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 Meta: Llama 3.3 70B Instruct that’s currently DeepInfra at $0.10/1M.
  • Batch tier — async, biggest discount. Roadmapped to use provider batch APIs (OpenAI / Anthropic) where available and queued spot capacity for open-weight workloads.