AI Model ID Cheat Sheet — the exact API model strings

This is a reference lookup, not a calculator — so “methodology” here means data provenance and the transform from your filters to the rows you see. The data lives in a single typed module, lib/data/ai-model-id-cheatsheet.ts, holding 29 models across seven providers.

1. Transcription, not inference. Every modelId, alias and endpoint is copied verbatim from the provider's official documentation. Each row stores its own source URL, surfaced as the “Source” link on every card.
2. Aliased vs pinned. Where a provider offers both an undated alias (gpt-4o, claude-haiku-4-5) and a dated snapshot (gpt-4o-2024-08-06, claude-haiku-4-5-20251001), both are stored. The ID-variant toggle picks which one the pill and snippet show. Pinned snapshots are reproducible; aliases track the newest version, which can shift behaviour — so production code should usually pin.
3. Filtering. The displayed rows are simply models.filter(provider && modality && nameMatches). It is pure and synchronous: chips narrow by provider and modality, the search box does a trimmed, case-insensitive substring match over the display name, model ID, alias and provider. No chip selected means “show all”, not “show none”.
4. Snippet generation. For a model, language and variant, the snippet is a static template with the resolved model ID and endpoint interpolated in. Anthropic rows use the anthropic SDK shape, Gemini rows the google-genai shape, and the OpenAI-compatible providers (OpenAI, Mistral, DeepSeek, xAI, Meta's Llama API) use the openai client with the right base URL. Keys are always YOUR_API_KEY placeholders — nothing executes, nothing leaves your browser.
5. Integrity check.A cross-check function validates that every stored ID matches its provider's documented naming pattern (Anthropic IDs start with claude-, xAI with grok-, and so on). The card footer reports the pass count, so a future mistyped entry is caught rather than silently shipped.

There are no rates or formulas to get wrong — correctness reduces to faithful transcription, which is exactly why the per-row source link and the LAST_VERIFIED date (Jun 29, 2026) are mandatory. For context windows, pricing and knowledge-cutoff dates, this page deliberately links out to the dedicated tools rather than duplicating them.

• OpenAI — Models (IDs, snapshots, endpoints)
• Anthropic — Claude models overview
• Google — Gemini API models
• Mistral — Models overview
• DeepSeek — Models & pricing
• Meta — Llama documentation
• xAI — Grok models

All model IDs, aliases and endpoints were last cross-checked against these provider docs on Jun 29, 2026. Provider model lists change often; the page is reviewed on a rolling basis and each card links to its source so you can confirm the live value.