Reasoning models get cheaper, AI agents move into spreadsheets and docs, and local multimodal inference takes a big step forward.
OpenAI shipped GPT-5.1 with automatic reasoning-depth scaling — it thinks longer only on hard prompts, cutting average serving cost 35 percent.
Anthropic released a native Excel add-in where Claude reads, edits, and audits live spreadsheets with cell-level citations.
Stability AI released SD 4.5 weights under a permissive community license, with best-in-class prompt adherence at 8B scale.
Scheduled agents now draft weekly reports, triage feedback databases, and update project pages autonomously inside Notion.
Luma AI shipped Ray3 with true HDR output and keyframe-consistent multi-shot scenes, targeting professional color pipelines.
Meta launched first-party hosted inference for Llama models with one-click fine-tuning, closing the gap with commercial API providers.
Gemini Live can now reason over a continuous mobile camera feed with sub-second responses for translation, repair help, and navigation.
Full-length tracks with vocals, structure control, and per-stem downloads arrived in beta for all paid tiers.
OpenAI open-sourced Whisper v4 with a 28 percent WER reduction on noisy, accented speech plus streaming transcription support.
The new cards ship with 32GB VRAM at mainstream pricing, explicitly targeting local 70B-class LLM inference for enthusiasts.
New editions are published every Sunday morning. Subscribe to our free newsletter or feed to never miss a breakthrough.
