Llama 2 70B Chat — Available Providers
Last refreshed 2026-05-16. Next refresh: weekly.
Compare pricing and deployment options across 14 providers.
Monthly cost ranking
Ranked for 1M input and 0.2M output tokens per month. Cache and batch discounts are applied only when the provider row has sourced prices.
Traffic profile
US default
Cost bars
Cheapest
Lepton AI API
$0.60 at the selected traffic profile.
Recipe-ready
DeepInfra
Install, auth, and call snippets are available from curated provider snippet data.
Provider matrix
Region, SLA, and compliance rows are hidden until curated source fields exist.| Provider | Profile cost | Input / Output | Deploy | Recipe | Links |
|---|---|---|---|---|---|
| Lepton AI API inference | $0.60 | $0.50 / $0.50 | Serverless | Docs only | |
| DeepInfra inference | $0.77 | $0.64 / $0.64 | Serverless | Snippets | |
| Databricks Foundation Model Serving platform | $0.80 | $0.50 / $1.50 | Serverless | Docs only | |
| Fireworks AI inference | $1.08 | $0.90 / $0.90 | Serverless | Snippets | |
| Together AI inference | $1.08 | $0.90 / $0.90 | Serverless | Snippets | |
| Replicate API marketplace | $1.20 | $0.65 / $2.75 | Serverless | Snippets | |
| GCP Vertex AI hyperscaler | $1.28 | $0.80 / $2.40 | Serverless | Snippets | |
| Microsoft Foundry hyperscaler | $1.89 | $1.54 / $1.77 | ServerlessProvisioned | Docs only | |
| IBM watsonx platform | $2.16 | $1.80 / $1.80 | Serverless | Docs only | |
| AWS Bedrock hyperscaler | $2.46 | $1.95 / $2.56 | Serverless | Snippets | |
| Alibaba Cloud PAI-EAS platform | Not enough pricing | - / - | Serverless | Docs only | |
| NVIDIA NIM inference | Not enough pricing | - / - | Provisioned | Docs only | |
| OCI Generative AI hyperscaler | Not enough pricing | - / - | Serverless | Docs only | |
| Scale AI GenAI Platform platform | Not enough pricing | - / - | Serverless | Docs only |
Llama 2 70B Chat operational data note: this page ranks sourced token, cache, and batch fields only. Region, SLA, compliance, and latency claims are intentionally omitted until a curated matrix is added.