✦AIToolsEssentialsPremium

Trust layer · return before you buy

What changed since the last check?

A transparent radar for recorded pricing and model-lineup freshness. It is not a vendor API or a guarantee of real-time change detection—always verify the official source before purchase.

Open Pricing WatchInspect Evidence LedgerKeep/Cut Weekly
Recorded, not impliedDates show when the source was last checked.
Models move tooLineup checks are separate from pricing checks.
Unresolved stays visibleMissing evidence is a buying signal, not a hidden footnote.
Vendor launches · not directory SKUs

Recorded launches that are not directory SKUs.

These are dated official-source watches. They are not tools.json listings and have no invented plan prices. Early-access launches stay here until public self-serve pricing exists. Open-source engines stay here unless they meet the directory listing bar.

Unlisted launch · limited rollout

GPT-6 Astra (OpenAI)

Checked 2026-09-03 against Path to Astra: critical capabilities and frontier safeguards.

OpenAI released GPT-6 Astra on 2026-09-03. The official Path to Astra post (2026-09-01) is still the primary openai.com page for the model as of this check: Astra is OpenAI's first model designated at Critical cybersecurity capability under the Preparedness Framework. Advanced cyber workflows are limited at launch to trusted testers / Daybreak Blue. Axios reports initial access for limited Daybreak organizations, then ChatGPT Plus / Pro / Business / Enterprise and API developers in the coming days. OpenAI customer stories dated 2026-09-03 use the name GPT-6 Astra. Greg Brockman called it a generational leap. This watch does not treat AGI remarks as fact.

Pricing: Roll-out onto existing ChatGPT paid tiers (Plus, Pro, Business, Enterprise) and the API was announced for the coming days. Sources checked 2026-09-03 did not confirm a Plus/Pro plan price change or a new Astra-only SKU. Advanced cybersecurity access is gated separately to trusted testers / Daybreak Blue. ChatGPT dollar prices were not re-verified on this date.

Monthly digest →Model lineups →Pricing Watch →Local AI planner →

Unlisted launch · early access

Atlas (World Labs)

Checked 2026-09-01 against Atlas: A World Model for Spatial Intelligence.

World Labs' next-generation omni world model, pretrained from scratch on text, images, video, and 3D. A multimodal autoregressive diffusion transformer with spatial context. Camera-controlled image/video generation (up to 1 minute at 1440p), spatial reconstruction from 1–dozens of images with explicit 3D (point clouds, 3D Gaussian splats), space-time simulation including video reframing and Real-to-Sim for robotics, plus text-to-image and 360 panoramas. Will power future versions of Marble and other World Labs products.

Pricing: Early access with select partners. Request access on the official blog. Not a public self-serve product; no published plan prices as of 2026-09-01.

Monthly digest →Model lineups →Pricing Watch →Local AI planner →

Unlisted launch · open source

Lily (Perplexity)

Checked 2026-09-02 against Optimizing On-Device Inference for Apple Silicon.

Official Perplexity blog (checked 2026-09-02): Lily is a lightweight local inference engine built specifically for Apple silicon and Qwen3.6-35B-A3B, with separate prefill and decode optimizations, so on-device compute does not bottleneck Hybrid Compute tasks in Perplexity Computer. A Rust runtime loads the checkpoint and runs the generation loop; an OpenAI-compatible chat-completions API accepts requests; custom Metal kernels execute Qwen-specific operations. Neither PyTorch nor MLX is in the execution path. A standalone demo is on GitHub. The official pplx-garden README names that demo at github.com/perplexityai/pplx-garden (lily/): Apache-2.0 Metal inference server, greedy generation, Qwen3.6-35B-A3B MLX affine 4-bit checkpoint format, Apple GPU family 10 or later (M5 and newer), macOS 26 or later. The garden README describes a minimal OpenAI-compatible HTTP API; the lily README as of this check documents POST /v1/chat/completions, GET /v1/models, and GET /health, and says streaming, sampling, tools, and multimodal requests are rejected. Vendor-published throughput figures were not independently reproduced.

Pricing: Open-source Apache-2.0 demo in perplexityai/pplx-garden (lily/). Not a Perplexity Pro/Max/Enterprise SKU. No public Lily plan price. Hardware, model weights, and any Perplexity Computer subscription are separate and were not priced on this check.

Monthly digest →Model lineups →Pricing Watch →Official demo →Local AI planner →

75 tools shown
Next decision step

Do not buy from a stale snapshot.

Use the radar to find the right record, then run the Fit Interview, inspect the evidence row, and test the tool with one real task before subscribing.

Run Fit InterviewCreate Decision Brief