The Platform PM

Glossary and sources

Field Guide

Models and inference: what AI runs on

10

Glossary and sources

7 min read · Chapter 10 of 10

Words that mean something else here

TermWhat you'd assumeWhat it means here
TokenA wordA piece of text in one model's tokenizer; the same text can be about a third more tokens on a newer model
LatencyOne numberTime to the first token (for reasoning models, the first thinking token), time to the first visible token, or end to end
ThroughputSpeedTokens a second one user sees, or tokens a second a GPU produces; they trade against each other
Context windowWhat the model can useThe maximum input and output; quality falls well before it
PrioritySame thing everywhereCommitted capacity (Anthropic, no longer sold), a 2x pay-as-you-go tier (OpenAI, now "Fast"), 1.8x per request (Google), +75% (Bedrock)
Fast modeA faster modelThe same model served faster at about twice the price; at OpenAI, the renamed priority tier
DeprecatedGoneStill works, with a retirement date set; in Azure's API, the value Deprecated means retired
Open sourceOpenUsually open weights under a licence, without training data or code; the AI Act's open-source exemption needs a free licence and no monetization
ProviderA vendorAn inference company, or the AI Act's party that places a model on the EU market, which a heavy fine-tuner can become
UptimeThe model worksThe endpoint accepts requests; nothing about the quality of the tokens
Run-rateRevenueRecent revenue times twelve; usage businesses aren't contracted for it
GigawattChipsPower capacity in a compute deal; contracted, energized and in use are three different numbers

What surprised me

Placeholders in your voice, drafted from the research and the earlier guides. Rewrite each with your own moment.

The price sheet. In telco a minute had a price and a carrier sold it. Here the price per token fell every year and the same feature could still cost more, because the model started thinking before it answered.

The rate limit. In observability I thought of 429s as a client bug. Here they're how a supplier with too few GPUs decides who gets served.

The SLA. In cloud security an outage was the worst case. Here the model can stay up and quietly answer worse for weeks, and no contract I found counts that.

The supplier. In the AI coding agents guide a lab could cut off a rival. Here the chip maker, the cloud and the lab invest in one another, so the supplier is also the customer's shareholder.

Sources

Undated entries were read on October 9, 2026; "search result" means seen only as a search snippet or summary. Company figures are self-reported unless they come from a filing or a regulator. Lab revenue and margin figures from leaks are unaudited.

Labs and clouds: documentation, pricing and terms

Company filings and disclosures

Labs' finances and deals (press and leaks)

Inference providers, GPU clouds and gateways (press)

Research, benchmarks and telemetry

Law, regulators and courts

Incidents

Field Guides are learning notes, not legal or compliance advice. Rules and fees change; check the cited primary sources before you act on anything here.