Gemini 3.7 Flash, Writer cuts agent costs: Aug. 14

Gemini just made repeated AI work cheaper, Writer added hard spending controls, Google Sheets can turn a table into a working dashboard, and Suno now gives you finer control over campaign music. Give this dispatch 30 minutes. You'll leave with one model test, a firm budget cap, a reporting prototype, and a safer audio process.

Gemini 3.7 Flash halves its introductory model price

Google released Gemini 3.7 Flash on August 13 at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens through December 31. Tokens are small pieces of text the system reads and writes. Google says that price is half the original Gemini 3.6 Flash rate, while its own AutomationBench score rose from 17% to 30.4%.

That combination matters when an AI step runs hundreds of times, such as sorting enquiries or summarizing call notes. A cheaper model still costs more when poor answers force repeat runs or staff corrections.

Open the workflow that consumes the most AI text. Duplicate it, change only the model to gemini-3.7-flash, and feed both versions 20 past jobs whose correct results you know. Record pass, correction, and failure counts beside the token cost. The AI prompt testing system gives you a plain checklist if you do not already have one.

Google's Gemini 3.7 Flash announcement says the model is available through its software connection and inside Gemini Spark. Without a technical workflow, test the repeated task in the interface you already use.

Writer adds Palmyra X6 and hard spending limits

Writer launched Palmyra X6 on August 13 at $2 per million input tokens and $8 per million output tokens. Writer also added real-time usage reporting, spending alerts, and a control that blocks further consumption above a set limit. Its agent, software that can complete several steps toward a goal, can now run one job for up to eight hours.

The useful release is the stop switch. Long jobs can repeat a bad instruction or call paid tools. Writer's figures come from its own testing, so run a controlled comparison.

If your team uses Writer Agent, open AI Studio > Home, filter usage by team or billing group, and note last month's actual consumption. Set an alert at 80% of your approved monthly amount and a hard block at 100%. Then assign expensive models only to jobs where better judgment changes the outcome. This is the same one-job-per-tool discipline in the small-team AI stack.

Writer's August product announcement says the controls and model are available now. If you do not use Writer, copy the policy: every automated AI account needs an owner, monthly amount, and shutoff point.

Google Sheets turns rows into an interactive canvas

Google began rolling out Sheets canvas on August 13 to eligible Workspace and Google AI plans. A canvas is a visual screen built over the spreadsheet. Changes made on that screen update the underlying cells, and cell edits update the screen in real time.

This can turn an approved scorecard into a dashboard. Keep the spreadsheet as the source of truth. The canvas should make approved numbers easier to read, not calculate a second version.

Open the scorecard, select Ask Gemini, then Create canvas. Prompt it to show the current week, the previous week, and a warning beside any blank required field. Check every displayed number against the cells before sharing. Google's Sheets canvas announcement says the feature is English-only at launch and lives as a shareable tab.

Start with the table in the weekly marketing report build. Do not replace its Approved release field. A polished screen built on an unfinished row is still a bad report.

Suno Studio 2.0 adds editable notes and custom effects

Suno released Studio 2.0 on August 13 with MIDI editing, a chat bar, improved stem separation, custom effects, and automation curves. MIDI is note-by-note performance information rather than recorded sound. A stem is one isolated part of a track, such as drums or vocals.

For campaign audio, the change is control. You can adjust the notes, separate an imported track, and draw volume changes instead of regenerating a whole song because one section is wrong. But Suno says a future credit structure may apply to custom plugins, so do not price a recurring production process on today's zero-credit rule.

Create a 15-second test, separate its parts, lower the vocal or lead part under spoken copy, and export the final mix plus its individual stems. Record the source project, prompt, editor, export date, and intended channels in your asset log. Suno's Studio 2.0 announcement says the release is currently tied to its Premier plan.

On the bench

• OpenAI's Ultrafast GPT-5.6 Sol mode promises up to 14 times Standard speed, but access is limited and no public price is posted. Wait for billable terms.
• DeepSeek Harness opens its agent machinery as replaceable plugins. Test it only in a copied workspace because the developer preview warns that its connections will change.
• Microsoft is combining separate Copilot experiences. Inventory saved instructions before any interface migration reaches your account.

Get the next build from the AI tools runbooks.