Elorian Glimpse: visual-thinking model that rewinds images/video mid-reason
At a glance
Elorian (Oct 16) launches Glimpse — a visual-thinking model that revisits evidence mid-reasoning (zoom, denser frame rewatch, identity tracking) and can spawn multimodal subagents. Vendor evals: ahead of GPT-6 Astra on several chart/counting/video tasks; API priced at $0.60/M input and $3/M output (~94% below Astra on those rates). Built by a small team on a few hundred GPUs atop an open-weight base; API + chat UI live. This workflow is listed under vision-model, robotics, video-reasoning, API, launch.
RAX (Robot Agent eXploration) lets a frozen LLM coding agent drive real robots via APIs, explore online, and compile successful perception–planning–control loops into reusable programs—no teleop demos or policy training; +37.4% avg vs baselines in sim; open skill libraries per embodiment.
Anthropic took Opus 5.5 fast mode out of research preview: set `speed: "fast"` (no beta header) for up to 2.5× OTPS at $8/$40 per MTok; separate rate limits; Claude Code `/fast` for interactive loops.
Kuaishou Kling AI fully rolled out Kling 4.0 (Oct 15): native 3–30s single-pass video, up to 10 keyframes, Omni Reference with ~15 multimodal inputs, smoother motion/camera control. Creation UI refreshed; developer path includes MCP and CLI so coding agents can drive generation without only the web UI. Feature page + official IR release; native 4K/HDR and broader enterprise API still listed as coming soon on some coverage.