Your AI strategy just got repriced.

Google cut the cost of its newest Flash model in half. SpaceXAI made Grok smarter without raising the price. OpenAI made its most powerful model run up to 14X faster.

That matters way more than another leaderboard win.

Price and speed are what separate flashy AI demos from workflows your business can actually afford to run every day. When both move this fast, use cases you ruled out a month ago deserve another look.

And that was just the model news.

Claude can now work inside your logged-in browser tabs. Google Sheets can turn live data into mini-apps. ChatGPT can watch how you work on your Mac, remember it, and help pick up what you forgot.

If your AI roadmap is based on last month’s assumptions, this week’s Friday Features might change it.

Google released Gemini 3.7 Flash just three weeks after 3.6 Flash, with gains in coding, debugging, and building web apps.

Paid Gemini users won’t find it in the normal model picker yet. It powers Gemini Spark and is available through Google AI Studio, Android Studio, the Gemini API, and Gemini Enterprise.

The bigger story is price. API pricing dropped to $0.75 per million input tokens and $3.75 per million output tokens, half the original 3.6 Flash cost.

For businesses running AI at serious volume, that can change the economics of an entire workflow. Something that looked too expensive to scale last month might suddenly make a whole lot more sense.

Try This

Take one high-volume AI workflow and rerun the cost math using Gemini 3.7 Flash.

If the quality holds while your token bill drops, you may have found a new workhorse model.

Z.ai released GLM-5.3, using the same underlying model as GLM-5.2 but improving it through better post-training.

Z.ai says it is roughly 50% better at coding on its own tests, scored 84.5 on CyberGym, and has found more than 1,000 critical security flaws in real software.

GLM Coding Plan and Z Code subscribers can use it now. The open weights are expected in about two weeks after additional safety testing.

GLM was already a budget favorite for AI coding. Now the cheaper option gets meaningfully stronger at complex, long-running work.

Try This

Rerun one of your nastiest long-running coding tasks on GLM-5.3 this week.

If the budget model handles it cleanly, rethink where you actually need the most expensive frontier option.

SpaceXAI released Grok 4.6 for coding and long-running automated work, jumping five points over the previous Grok on Artificial Analysis.

Developers can access it through the xAI API, it is included on all Cursor plans, and it is now the default model in Grok Build.

The price stayed the same as Grok 4.5.

That keeps the model price war moving. More intelligence without more cost means Grok deserves another look from businesses that previously wrote it off.

Try This

Run one familiar coding or agent workflow through Grok 4.6 and compare it with your current default.

If the output is competitive at the same price, update your shortlist instead of relying on old assumptions.

Google expanded Sheets canvas, letting you turn a normal spreadsheet into an interactive mini-app or dashboard that stays synced with the underlying data.

Paid Gemini subscribers can access it from the Gemini panel inside Sheets, and it is also available for Gemini work and school accounts.

The unlock is not prettier spreadsheets. It is taking data your business already updates inside Sheets and turning it into something your team can actually use without rebuilding the view every week.

If your CRM, ads, email, or other systems already feed data into Sheets, you can build the app once and let the source data keep it current.

Try This

Pick one Sheet your team checks constantly and ask Gemini to turn it into the dashboard you actually wish existed.

If the data already syncs automatically, you just removed another recurring reporting chore.

Claude in Chrome’s side panel is now a Claude Cowork session, so Claude can see the page you’re on, click, type, fill forms, and use your existing logins.

Tasks can move between the browser and Cowork across Claude’s desktop, web, and mobile apps. Max and Team plans have access now, with Pro rolling out over the coming weeks.

This matters most for the giant chunk of business software that still has no clean AI connector or MCP.

Vendor portals. Internal dashboards. Older web apps. If Claude can see and use the interface, lack of an integration no longer automatically kills the workflow.

Try This

Pick one annoying browser-based tool your team uses that has no AI integration and give Claude one repeatable task inside it.

If it can reliably work through the interface, you may have just unlocked automation without waiting on the software vendor.

OpenAI launched Ultrafast, a new API speed tier that runs GPT-5.6 Sol up to 14X faster through a partnership with Cerebras.

It can generate up to 750 tokens per second, with OpenAI and Cerebras saying it delivers the same intelligence at dramatically higher speed. Access is currently in limited preview.

Until now, getting real-time speed usually meant dropping down to a smaller, less capable model.

That tradeoff just got weaker. On a long agent workflow, shaving time from every step can turn something that takes minutes into something that finishes in seconds.

Try This

If your company has access, test Ultrafast on one customer-facing or multi-step workflow where latency is currently painful.

If waiting is the only thing keeping that use case out of production, the equation just changed.

OpenAI’s Computer History watches activity across apps and websites on your Mac, then turns it into memories and a timeline that ChatGPT and Codex can use.

It is Mac only, off by default, and available to ChatGPT Pro and paid Business plans. It is not yet available in the EU, UK, or Switzerland.

The value goes way beyond asking what you worked on yesterday.

Computer History can help resume abandoned work, surface where you left off, and suggest reusable skills based on repeated workflows. For anyone bouncing between a pile of projects every day, that directly attacks the silent tax of context switching.

Try This

Turn it on, then end your day by asking ChatGPT to summarize what you worked on and start tomorrow by asking it where you left off.

If it starts spotting repeated workflows you can save as skills, that is where this feature gets really interesting.