RADAR ARCHIVE

Week 29, 2026

3 tools added to the directory and 5 launches tracked, Jul 13 – 19, 2026.

Added to the directory (3)

Tested and written up.

Moonshot AI's 2.8-trillion-parameter open-weight flagship
Added Jul 16, 2026
Kimi K3 is Moonshot AI's open-weight model released July 16, 2026, built at roughly 2.8 trillion parameters. It's available now via kimi.com, Kimi Work, Kimi Code, and the Kimi API, priced at $0.30 per million input tokens (cached), $3.00 per million (uncached), and $15.00 per million output tokens. Moonshot published the full open weights on Hugging Face on July 27, 2026, as promised at launch. Moonshot positions it as approaching Anthropic Fable 5-tier performance at a fraction of the cost, though the company acknowledges a tendency toward 'excessive proactivity' on long-running tasks.
Why: Kimi K3 is one of the most credible open-weight challengers to closed frontier models this year, aggressive enough on pricing and scale that it moved markets, Fortune covered it as a 'DeepSeek shock' moment for AI stocks.
Freemium Best for Open-Weight Frontier Performance Visit
Controllable cinematic video model with multi-keyframe direction and motion transfer
Added Jul 15, 2026
Luma Ray 3.2 is a production-grade video generation and editing model that creates 1080p clips up to 20 seconds from text, images, or existing video. It supports up to 16 multi-keyframes, motion and camera transfer, character transformation, environment changes, relighting, and native HDR/EXR export for post-production workflows.
Why: Ray 3.2 gives professional teams frame-level control over video generation, including motion transfer and EXR export, making it a strong contender for production pipelines.
Freemium Best for Cinematic Control Visit
Alibaba's 2.4-trillion-parameter flagship, currently in preview
Added Jul 19, 2026
Qwen 3.8-Max is Alibaba's largest model yet, announced July 19, 2026, at 2.4 trillion total parameters using a sparse Mixture-of-Experts design. It's multimodal (text, images, video, documents) with a context window in the ~1M-token range (983,616 tokens per Qwen Cloud metadata) and a 131,072-token max output. It's live now as qwen3.8-max-preview through Alibaba's Token Plan, Qoder, and QoderWork at 10% of eventual standard pricing, targeting coding, agentic workflows, and long-horizon 'professional cowork' tasks. Alibaba says open weights are coming but hasn't published a date, license, model card, or full benchmark table yet; the independent number available is Artificial Analysis, which places the preview at 53.4 on its Intelligence Index, rank 11.
Why: Qwen 3.8-Max is Alibaba's answer to the current wave of massive open-weight-adjacent models from Chinese labs, and the discounted preview pricing makes it worth evaluating early even before the full release details land.
Paid Best for Long-Horizon Agentic Work (Preview) Visit

On the radar (5)

Reported in the daily briefing this week. Not reviewed for the directory.