World models + robotics cluster: World Labs RTFM + DeepMind Genie 3 + NVIDIA Cosmos + Figure Helix 2 — every world model / humanoid robotics vendor ships a GA milestone in one week

6 sources

The week of 2026-08-11 saw a cluster of world model + robotics releases: World Labs RTFM (real-time feed-forward 3D generation from a single image, no per-scene optimization), DeepMind Genie 3 GA (interactive world model — user can navigate + interact in real time), NVIDIA Cosmos (world foundation model for robotics + autonomous systems with Cosmos Reason VLA + Cosmos Transfer simulation-to-real), Figure Helix 2 (vision-language-action model for humanoid robots with dual-system architecture). The Decoder frames the cluster as 'every world model + humanoid robotics vendor ships a GA milestone in the same week'. The pattern: world models are converging on real-time + interactive output (RTFM + Genie 3) and robotics foundation models are converging on VLA architectures (Cosmos + Helix 2).

Why now

The cluster is the editorial framing that turns four independent world model + robotics releases into a single coherent 'world models and VLA robotics hit GA simultaneously' story for creators.

Why it is worth publishing

Demo potential: side-by-side of a real-time 3D scene generated by RTFM and navigated via Genie 3, plus a side-by-side of a Cosmos-driven simulation task vs a Helix 2 humanoid task.

Evidence basis

The Decoder + IT之家 weekly roundups + four independent vendor primary sources

“Four world model + humanoid robotics vendors shipped GA in one week — World Labs RTFM, DeepMind Genie 3, NVIDIA Cosmos, Figure Helix 2 — and the world models + robotics foundation models just hit creator-researchable surfaces simultaneously.”

Angle

Use the cluster to introduce the 'world models + VLA robotics hit GA simultaneously' pattern and use that lens to compare approaches side-by-side.

Format

Long-form explainer

Demo idea

Record a 14-minute comparison explainer: 2 min intro on 'world models + VLA robotics hit GA simultaneously' framing, then 2 min per vendor (RTFM / Genie 3 / Cosmos / Helix 2), then a 4-min side-by-side of a real-time 3D scene + a VLA robotics task across the four.

Platform notes

Each vendor's release notes frame its release against the competitive set it cares about; The Decoder and IT之家 are editorial framing layers, not independent verification. Confirm any specific latency target or robotics benchmark against the underlying vendor docs before stating it on the record.

Usable claims

World Labs shipped RTFM — real-time feed-forward 3D world model that generates 3D scenes from a single image without per-scene optimization.
Google DeepMind shipped Genie 3 GA — interactive world model that lets a user navigate and interact with a generated environment in real time.
NVIDIA shipped Cosmos — world foundation model for robotics + autonomous systems, with the Cosmos Reason vision-language-action model and the Cosmos Transfer simulation-to-real pipeline.
Figure shipped Helix 2 — vision-language-action model for the Helix humanoid robot, with the dual-system architecture (System 1 fast reactive + System 2 slow reasoning).

Evidence pipeline

From the news

Breakdown

Four world model + humanoid robotics vendors shipped GA in one week — the editorial framing ('world models and VLA robotics hit GA simultaneously') is useful but turns into a research lab tour if you don't introduce the convergence pattern. This explainer uses the cluster to introduce the convergence pattern (world models on real-time + interactive, robotics foundation models on VLA) and uses that lens to compare approaches side-by-side.

Sources

Risks

Use The Decoder and IT之家 as media-type corroboration, but read the underlying vendor docs for any specific capability or benchmark claim before stating it on the record. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.
Vendor docs confirm real-time / interactive capability but specific latency targets beyond the captured summary are not in this pass. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.
Vendor docs confirm feature existence but specific benchmark numbers beyond the captured summary are not in this pass. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.

Demo ideas

Side-by-side real-time 3D + interactive: same scene generated by RTFM and navigated via Genie 3.
Side-by-side VLA robotics: Cosmos-driven simulation task vs Helix 2 humanoid task.
Decision tree: 'which world model / VLA for which use case' (real-time 3D generation → RTFM, interactive navigation → Genie 3, general-purpose robotics → Cosmos, humanoid-specific → Helix 2).