Ling-3.1-flash: a 560B MoE, free on OpenRouter right now
inclusionAI's hybrid-reasoning MoE gives you 256K context at zero cost for now, open weights promised later - plus Kling 4.0 Flash's 30-second 4K video.
Copy markdown
A frontier-scale MoE, free to call right now
inclusionAI (Ant Group's open-model lab) put Ling-3.1-flash live on OpenRouter via NovitaAI - 560B total params, 25B active, a 262K-token context, and $0 pricing for now - so you can benchmark a frontier-scale reasoning model on your own tasks at zero API cost. The “open weights” are only promised for after the trial (1M context at full release), so it's API-only today; evaluate against it, but don't plan on self-hosting yet.
Kling 4.0 Flash: 30-second 4K video, HDR and stereo
Kuaishou's Kling shipped 4.0 Flash to annual (Ultra Yearly) subscribers ahead of a public release this month - up to 30-second clips at 4K 10-bit HDR with stereo audio and 10 keyframes for shot direction. If you're building short-form video tools, the ceiling on length and fidelity just jumped; broader API access should follow the public rollout.