docs: refresh the model fleet and link the chat / code / design surfaces

[AetherAI3]

What

The README's model copy predated two AetherCloud releases. It listed five providers and "image and video generation" with no detail — nothing told a reader the frontier tier exists, and the --model example pointed at opus.

Models & pricing

• Frontier table (Pro/Team) — the six models that shipped July 25, each with its exact --model key and context window: Claude Opus 5 (opus5), GPT-5.6 Sol / Terra / Luna (gpt56_sol · gpt56_terra · gpt56_luna), Kimi K3 (kimi_k3), Gemini 3.6 Flash (gemini36_flash).
• Opus 5 is called out as additive — --model opus still resolves to Claude Opus 4.8. Anyone with it pinned reads that and stops worrying.
• Per-plan table for the rest of the fleet (Free / Solo / Pro-Team), so the gating is visible without signing in.
• Media expansion from the July 27 release: 15 image + 16 video + text-to-3D, up from the original eight.
• --model flag example moves to real frontier keys.

Platform links

Part of the Aether platform mentioned only the desktop app. It now links every surface, with a line each on what it is:

Surface	Link
Aether AI on the web (the Workbench)	https://app.aethersystems.net/chat
Aether Code	https://app.aethersystems.net/code
Aether Design	https://app.aethersystems.net/design
AetherCloud desktop	https://github.com/DBarr3/aethercloud
Aether Terminal	https://aethersystems.net/terminal

RELEASE_NOTES.md

Gains an August 6 fleet entry covering the same ground, framed correctly — the fleet is server-side, so an installed CLI already reaches these models with no client upgrade. It also corrects the June note: that release shipped vision_kling as Kling 3.0 Standard; the "Kling 2.5 Turbo Pro" label named a model the backend does not call.

Provenance

Every model name, key, context window and tier gate is read from AETHER-CLOUD origin/main — lib/model_registry.py, lib/plan_tiers.py, lib/model_catalog.py — not from the release-notes prose. Keys with enabled=False (kimi_k26_thinking, gemma4_31b_free, gemma4_26b_moe) are deliberately not advertised, since they render as locked rows.

All five URLs verified 200.

Not in this PR

src/ui/model_picker.ts maps provider google → the label "Gemma". With Gemini 3.6 Flash now in the fleet, that group label is wrong for one of its two models. Left alone here — it's a code change, not a docs one.

🤖 Generated with Claude Code