Guides·Inference

Inference providers

Model weights are increasingly a commodity; serving them fast and cheap is not. These guides compare the providers on the numbers that matter for agentic coding: price per million, speed, and reliability.

Download for Mac All guides
01

Start here

Inference

Inference providers: the honest 2026 comparison

Five kinds of company sell you inference. Here is what each actually charges, where each is genuinely fast, and which one a coding agent should point at.

Read guide →8 min Inference

LLM inference speed: tokens per second is the wrong metric

Throughput, first-token latency, and end-to-end turn time are three different numbers. Agents are bound by the one nobody advertises.

Read guide →8 min Inference

Cheapest LLM API: the honest list, by capability class

Cheap is only meaningful within a capability class. Here are the real numbers in each, and the five traps that turn a cheap rate into a large bill.

Read guide →8 min
02

Every guide in this cluster

Inference providers: the honest 2026 comparison Five kinds of company sell you inference. Here is what each actually charges, where each is genuinely fast, and which one a coding agent should point at. 8 min Fireworks AI review: what it is good at, and what it is not Fireworks serves open weights fast and sells the speed as a separate SKU. Here is what that buys, and where the catalog stops. 7 min Fireworks AI pricing: every rate, tier, and free credit The full Fireworks rate card, read off the vendor pages in August 2026, plus what a real coding-agent month costs. 6 min Together AI review: the widest catalog, and what it costs you Together sells the broadest surface in open-model inference: tokens, clusters, training, sandboxes. Breadth is the pitch and also the risk. 6 min Together AI pricing: tokens, GPUs, batch, and the free tier Every Together rate card in one place, read off the vendor pages in August 2026, including the parts the marketing page skips. 6 min Baseten review: model APIs, dedicated deployments, and pricing Baseten is two products wearing one name: a small fast model API, and a serious platform for deploying your own model. Know which one you are buying. 6 min Modal review: serverless GPUs, and when a team needs them Modal is not an inference API. It is a Python-native way to run any code on a GPU and pay by the second. That difference decides whether it is right for you. 6 min Modal pricing: every GPU rate, free credits, and worked costs The complete Modal rate card, converted to hourly, plus the utilization point where renting a GPU outright becomes cheaper. 6 min Cheapest LLM API: the honest list, by capability class Cheap is only meaningful within a capability class. Here are the real numbers in each, and the five traps that turn a cheap rate into a large bill. 8 min Serverless inference: three different things with one name The word covers a token API, a GPU container, and a reserved endpoint. They have different prices, different failure modes, and different right answers. 6 min LLM inference speed: tokens per second is the wrong metric Throughput, first-token latency, and end-to-end turn time are three different numbers. Agents are bound by the one nobody advertises. 8 min Free LLM API: every genuinely free option, with the catch for each Six providers that will serve you tokens for nothing, the exact limits, and what each one wants in return. 7 min
03

Other clusters

PricingUsage trackingInstallReferenceMCPCodexGeminiCursorWorktreesGrok BotHead to headConceptsTroubleshootingGatewaysSpend
Try it

Every agent.
One workbench.

Continuum runs Claude Code, Codex, Cursor, Gemini, and more under the subscriptions you already pay for, with live quota gauges and spend by repo.

Download for Mac Compare tools

free app · your subscriptions · local-first