RADAR ARCHIVE
Week 36, 2026
4 tools added to the directory, Aug 31 – Sep 6, 2026.
Added to the directory (4)
Tested and written up.
Google's AI-powered speech-to-text model
Gemini 3
Why: Google's entry into transcription space backed by Gemini's reasoning capabilities. Part of 3.5 release wave, positioning it alongside current-generation Google models. Useful for teams integrating speech-to-text into multimodal workflows.
Freemium
Best for Speech-to-Text
Visit
Open-weight video/world model, 10s clips from images in 6.8s
LTX-2
Why: LTX-2.5 dominates the open video/world model space (33M downloads). Open weights under permissive license, strong feature set (multi-shot, better quality, robotics support). For teams building video generation or world model workflows without proprietary constraints, this is the category leader.
Freemium
Best for Open Video Generation
Visit
30B on-device AI agent, runs natively on consumer hardware
Muse Glimmer is Meta's 30 billion parameter open-weight model released August 2026, designed to run autonomous AI agents directly on consumer hardware without cloud dependency
Why: Open weights on Apache 2.0, runs on consumer hardware, removes user-count restrictions that hampered Llama. For teams building local-first or edge-deployed agents, this eliminates licensing friction and eliminates inference costs.
Free
Best for On-Device Agents
Visit
118B MoE model beats rivals 10x its size on coding benchmarks
Laguna S 2
Why: Benchmark contender: claims to beat models many times its size on two critical coding benchmarks. Open weights under permissive license. Rapid training timeline (9 weeks) suggests efficient engineering. Strong SWE-Bench showing makes it worth evaluating for coding agent workloads.
Free
Best for Coding Performance
Visit