[02] Model provider · required/ What thinks

Model provider for AI agents

Where the tokens come from. A frontier API, a gateway you control, or your own GPUs.

13 options tracked · 3 open source · 3 self-hostable · required in every stack

· All 13 options/ Compare
OptionWhat it doesLicenceSelf-host
AnthropicClaude Opus, Sonnet and Haiku. Strong tool use and long-horizon agentic work.ProprietaryNo
OpenAIGPT and o-series via the Responses API.ProprietaryNo
Google GeminiGemini models with very long context and native multimodality.ProprietaryNo
Amazon BedrockFrontier models inside your AWS account, with IAM and VPC boundaries.ProprietaryNo
Google Vertex AIClaude and Gemini under GCP billing, IAM and regional controls.ProprietaryNo
DeepSeekStrong reasoning and coding at a fraction of frontier pricing, with the weights published so you can move off the API later.ProprietaryNo
OpenRouterOne API key, several hundred models, automatic failover.ProprietaryNo
Hugging FaceOne OpenAI-compatible endpoint routed across Groq, Together, Fireworks, Cerebras and Replicate — with the open-weight catalogue behind it.ProprietaryNo
Venice AIHosted open-weight inference with no prompt logging or retention, plus an anonymising proxy in front of the frontier APIs.ProprietaryNo
LiteLLMSelf-hosted proxy that speaks one API to 100+ providers, with keys and budgets.Open sourceYes
GroqOpen-weight models at very low latency.ProprietaryNo
OllamaOpen-weight models on your own machine. Nothing leaves the box.Open sourceYes
vLLMHigh-throughput open-weight serving on your own GPUs.Open sourceYes
· Published benchmark scores/ Every number sourced

Only figures with a published source appear here. A blank cell means nobody has published one for that pairing — which, for most of this category, is the honest answer.

BenchmarkAnthropicDeepSeek
SWE-bench VerifiedShare of real, human-validated GitHub issues resolved end to end.96%leaderboardClaude Opus 596.4%leaderboardDeepSeek V4 Pro 0813

SWE-bench Verified. Scores attach to a specific model, not to a provider — the model measured is named in each cell. Frontier results now cluster inside a single point and different leaderboards report different figures for the same model, so treat anything under ~1 point as noise rather than a ranking.

Sources
· Head to head/ 78 comparisons
· Common questions/ FAQ

What is the model provider layer of an AI agent?

Where the tokens come from. A frontier API, a gateway you control, or your own GPUs. Every stack needs one — it is not optional.

How many model provider options are there?

This registry tracks 13. 3 are open source and 3 can run on your own infrastructure.

Which model provider option should I choose?

It depends on constraints rather than preference: whether you must self-host, whether the budget allows a hosted service, and which language your team writes. Describe what you are building and the advisor fills this layer along with the other 9.

· The other 9 layers/ Keep going
· For agents/ This page, machine-readable

Every page here answers to Accept: text/markdown and returns the same content at roughly a tenth the tokens. No separate site, no toggle — same URL.

curl -s -H "Accept: text/markdown" https://newagent.build/layers/model