# API Atlas — OpenAI + Anthropic + xAI + Gemini

Exhaustive, experimentally verified reference of everything you can do with the OpenAI, Anthropic, xAI (Grok) and Google Gemini public APIs as of 2026-09-18/19: models, endpoints, parameters, objects, streaming events, tools, agents, realtime, media, files, batch, fine-tuning, evals, admin, pricing, rate limits, errors, headers, SDKs, security, and a multi-provider abstraction. Machine-readable first (generated/), human docs second (docs/), runnable examples (examples/) and smoke tests (tests/) third.

Every fact carries a status (DOCUMENTED, LIVE_VERIFIED, LIVE_DISCOVERED, BETA, PREVIEW, LEGACY, DEPRECATED, RETIRED, ACCOUNT_RESTRICTED, UNVERIFIED, FAILED_VERIFICATION) and its sources (official URL + retrieval date). Nothing is claimed "tested" unless it was actually called with a real key (see reports/live-requests.jsonl).

# Start here

Need	Go to
Overview & navigation	`docs/index.md`
Which model supports what	`docs/models/openai-models.md` · `docs/models/anthropic-models.md` · `docs/models/xai-models.md` · `docs/models/gemini-models.md` · `generated/compatibility/model-capability-matrix.csv`
Every endpoint	`generated/endpoints.csv` · `docs/endpoints/`
Every parameter	`generated/parameters.csv` (`generated/parameters.json` for nested detail)
Every tool (exact `type` strings)	`docs/tools/openai/index.md` · `docs/tools/anthropic/index.md` · `docs/tools/xai/index.md` · `docs/tools/gemini/index.md` · `generated/tools.json`
Streaming events (SSE / WebSocket)	`generated/streaming-events.json` · `docs/openai/streaming-events.md` · `docs/openai/realtime-events.md` · `docs/anthropic/streaming.md`
Agents platforms	`docs/openai/agents-api.md` · `docs/anthropic/managed-agents.md` · `docs/gemini/interactions-api.md` · `docs/xai/responses.md`
Pricing	`docs/openai/pricing.md` · `docs/anthropic/pricing.md` · `docs/xai/pricing.md` · `docs/gemini/pricing.md` · `generated/pricing.csv`
Errors & retries	`docs/errors/openai.md` · `docs/errors/anthropic.md` · `docs/errors/gemini.md` · `docs/xai/authentication-headers-errors.md` · `docs/architecture/resilience.md`
Cross-provider comparisons (4 providers)	`docs/comparisons/`
Security	`docs/security/README.md`
Multi-provider code	`docs/architecture/multi-provider-abstraction.md` · `examples/shared/`
What was verified, what is missing	`reports/final-report.md`

# Setup

bash

cp .env.example .env && chmod 600 .env   # add OPENAI_API_KEY / ANTHROPIC_API_KEY / XAI_API_KEY / GEMINI_API_KEY
uv venv .venv && uv pip install --python .venv/bin/python openai anthropic xai-sdk google-genai pytest pyyaml jsonschema websockets
pnpm install                              # openai + @anthropic-ai/sdk + @google/genai for the TypeScript examples

Run an example: bash examples/anthropic/messages/minimal.sh, .venv/bin/python examples/openai/responses/minimal.py, node --env-file=.env --experimental-strip-types examples/openai/responses/minimal.ts.

Run the smoke tests (cheap by default; expensive suites are gated by RUN_*_TESTS=true, see .env.example):

bash

.venv/bin/python -m pytest -q

# Re-running the atlas (change detection)

bash

python3 scripts/update_atlas.py          # refresh indexes + crawl docs + /v1/models + rebuild + reports/changes.md
python3 scripts/build_generated.py       # rebuild generated/ from generated/fragments/
.venv/bin/python scripts/validate_fragments.py
python3 scripts/verify_links.py          # external + internal link check → reports/link-check.json

reports/changes.md lists NEW/REMOVED MODELS, ENDPOINTS, TOOLS, PRICING CHANGES, CONTEXT WINDOW CHANGES, NEW BETA FEATURES, DEPRECATIONS and the official pages whose content changed since the last run.

# Layout

text

docs/          human documentation (openai/, anthropic/, xai/, gemini/, models/, tools/, endpoints/, errors/, comparisons/, architecture/, security/)
generated/     merged machine-readable outputs (+ fragments/ = curated inputs, + compatibility/ matrices, + capability graph)
examples/      runnable cURL / Python / TypeScript examples with verification status in each file header
tests/         pytest smoke tests (openai/, anthropic/, shared/)
scripts/       crawl, discover, build, validate, verify, update
sources/       downloaded official documentation (OpenAI 551, Anthropic 696, xAI 182, Gemini 291 pages), OpenAPI/discovery specs, SDK surfaces, sanitized live discovery
schemas/       JSON Schemas for every record type
reports/       live request log, coverage, changes, final report

Security rules: keys only in .env (600, gitignored), never in files or logs; admin endpoints are probed read-only; all live tests use minimal payloads. See CLAUDE.md for the full conventions.