GitHub and Lovable are quietly changing their terms today. Anthropic and Meta also shipped new models on the same day.
GitHub and Lovable are quietly changing their default terms this week.
Until now, seat counts adjusted with prorated billing as you added or removed people mid-month.
From October 2026, GitHub Copilot Business/Enterprise seats are billed upfront with no prorated refund when revoked
If you add or remove seats often, budget for the full seat cost upfront starting in October. Solo users on personal plans aren't affected.
Until now there was no clear default on whether your data trained the model, regardless of plan.
From September 9, Free/Pro usage data is used for AI training by default (opt out in Account settings)
If you build anything sensitive on Free/Pro, opt out before September 9. Business plans are opted out by default already.
Anthropic, Meta, and Mistral all shipped new models on the same day.
Until last month, Opus 5 was the top-tier model you could call directly.
Fable 5.1 joins the lineup, and the top tier is now Mythos 5.1 - no direct prompting, only authorized paths like Claude Security
Developers hitting the top model directly via API need to rework their access path. Regular Pro/Max users see no change.
The existing three-way lineup (Spark, Glimmer, Llama 4) had no dedicated transcription model.
A dedicated real-time transcription model, Muse Voice Transcribe, joins the lineup
Teams self-hosting meeting notes or captions get a new option; no change for text-generation use cases.
The specialist lineup so far covered OCR, code, speech, and safety - nothing for formal math proofs.
Leanstral 1.5, specialized in formal proofs, joins Mistral's task-specialist lineup
Useful for research or finance work needing theorem verification; unlikely to matter for everyday chat use.
Devin Desktop's Codemaps now runs on an in-house model, up to 950 tokens/sec.
Codemaps hadn't previously published anything about its processing speed.
Codemaps now runs on in-house model SWE-1.6, hitting up to 950 tokens/sec on paid plans
Engineers running monorepo or legacy-code analysis daily get a real speedup; casual solo use won't feel much difference.
A daily archive of changes actually applied to the site.
GPT-6 Astra is spreading to top plans and Gemini just picked up music generation. There are also 3 fresh usage tips today.
This morning's big news is flagship launches: GPT-6 Astra and Gemini 3.8 Flash. Agents are also starting to pick their own models and bro…
Google and Meta both refreshed their flagship models this morning. Copilot also picked up a dirt-cheap transcription model, plus 4 fresh …
GitHub and Lovable are quietly changing their terms today. Anthropic and Meta also shipped new models on the same day.
This morning ChatGPT lands under the EU's heaviest platform rules, and Runway stretches beyond video into UI generation, plus 5 quick tip…
Today's theme is agent governance and cost. Glean tightened agent permissions while Zendesk published its per-resolution overage price.
This morning: Anthropic officially announced Claude Sonnet 5, and ElevenLabs' voice conversations reached GA. Plus 3 quick usage tips.
Today's big moves: Gemini gets a dedicated video model and Anthropic faces a music copyright suit. Plus 6 usage tips, from Cursor tricks …
Today's update centers on Claude becoming reachable via Microsoft Foundry (Azure). There are also 4 new usage tips.
This morning Claude keeps burrowing into enterprise CRM and desktop apps. Voice transcription got smarter while ChatGPT's free tier picks…
Stability AI picked up big music-label money, and ChatGPT started poking around inside your open browser tabs. The image-gen pecking orde…
Perplexity had the day's only news: a fully local NVIDIA agent plus enterprise MDM rollout. Seven tips round things out.
Today: Manus restores deleted data and Anthropic locks down its flagship model. Plus five practical tips.
This morning's headline is Luma AI's Ray3.2, which lets you pin down exact video shots. There are also 5 usage tips today.
