THE ATLAS · v · LAST UPDATED · checking…

Every frontier AI model.
Text. Image. Video. Audio.

43 models. 17 vendors. One neutral table. Text reasoning, image generation, video, audio, voice cloning. Live pricing, context, capabilities, status — released vs. announced vs. expected, clearly marked. Auto-refreshes when new models ship — no reload needed.

Total models tracked
Shipping now (GA)
Announced + expected
Independent vendors

Status legend: GA shipping in production · ANNOUNCED vendor-confirmed, not yet GA · EXPECTED based on industry signals — preliminary. PR welcome at github.com/nexora-aero/atlas.

02 · COST PROJECTION · TEXT & MULTIMODAL

What will it actually cost you?

Token-based projection. Image / video / audio models price by image, second, or character — see their detail rows for unit prices.

03 · AUTO-UPDATE

This page refreshes itself.
No reload required.

The Atlas reads from /atlas/data.json on every load, with cache-busting. When a new model ships, we update the JSON — your tab gets fresh data within ~60 seconds of CDN propagation. While the tab stays open, we re-check every 10 minutes and animate in new rows.

Last fetch:
Data revision:
Source: /atlas/data.json · GitHub repo

LIVE UPDATE LOG
  1. 27 May 26 v2026.05.27.1 · added Imagen 4 Ultra, DALL·E 4, Sora 2, Veo 3, Suno v5, ElevenLabs v3, GPT-5.5, Claude Opus 4.7, Gemini 3.0, Llama 5
  2. 15 May 26 v2026.05.15.1 · added Magic LTM-3 (100M context)
  3. 02 May 26 v2026.05.02.1 · DeepSeek V3.5 launch pricing
  4. 18 Apr 26 v2026.04.18.1 · Initial release · 16 models
04 · FAQ

Questions procurement
actually asks.

If your answer isn't here, email atlas@nexoraaero.com.

How do "announced" and "expected" differ from GA?
GA = available right now, you can hit the API today. Announced = vendor has publicly committed to ship by a given quarter; pricing usually preliminary. Expected = strong industry signals (roadmap leaks, hiring patterns) but no public commitment yet. Use the "Released only" filter if procurement requires concrete availability.
How does the auto-update actually work?
The page ships with an embedded snapshot of the dataset (instant load, works offline). On load — and every 10 minutes thereafter — it fetches /atlas/data.json with cache: 'no-store'. If version is newer, the table re-renders with new rows highlighted briefly. To push an update, edit one JSON file.
Where does this data come from?
Directly from each vendor's public docs. We verify weekly. Every row links to its primary source. We do not accept vendor edits, sponsorships or paid placement.
Why aren't benchmark scores included?
Benchmarks are gameable. We publish objective measurable facts. For benchmarks we recommend artificialanalysis.ai.
Can I export this for procurement?
Yes. The raw dataset is at /atlas/data.json under CC0.
How do I avoid vendor lock-in?
Use the Constellation Protocol — open MIT-licensed standard for portable AI agents.
05 · PORTABLE BY DEFAULT

Pick a model. Then make it replaceable.

The Atlas tells you which model is best today. The Constellation Protocol makes switching to whichever model is best tomorrow a one-line config change.

Read the Constellation Protocol →