Three verified AI launches from Aug. 13: Ultrafast GPT-5.6 Sol, Gemini 3.7 Flash and Amazon Quick in Microsoft 365 (2026)

Aug. 13's meaningful launches split into three practical bets: make a frontier model answer faster, make a workhorse model cheaper to run, and put an agent inside the office tools people already use. The first is a limited preview; the other two are available now under specific access conditions.

Coverage window: Aug. 13–14, 2026, ending before this issue was prepared.

The scan at a glance

Date	What shipped	Builder impact	End-user impact	Access or price
Aug. 13	OpenAI previewed Ultrafast mode for GPT-5.6 Sol, with up to 750 output tokens per second and up to 14x the speed of Standard processing. 1	A faster API tier for synchronous support, research, incident response, and other workflows where waiting changes the product.	No direct consumer rollout is described; users may encounter it later through apps that adopt the tier.	Limited preview for a select group of customers. OpenAI is collecting access requests; no price is stated. 1
Aug. 13	Google released Gemini 3.7 Flash for coding and agent workflows, with a 1,048,576-token input limit and support for code execution, function calling, search grounding, and other tools. 2 3	A production API model with multimodal inputs, caching, batch and priority inference, and low/medium/high thinking controls.	Gemini Spark will use it for Google AI Pro and Ultra subscribers in supported countries. 2	Introductory API pricing is $0.75 per million input tokens and $3.75 per million output tokens through Dec. 31, 2026. The listed price then doubles. 2
Aug. 13	Amazon Quick became generally available inside Microsoft Word, Excel, PowerPoint, and Outlook through four extensions. 4	Teams can bring Quick's connected data, agentic editing, and existing integrations into Microsoft 365 without adopting another work surface.	Users can draft, edit, analyze, and act inside documents, spreadsheets, presentations, and email.	No additional license is required for existing Plus, Professional, or Enterprise customers. The extensions are available on desktop and web in seven AWS Regions. 4

Three bottlenecks, three different launches

OpenAI is selling time per answer

Ultrafast does not change the model's name or advertised intelligence. It changes the service tier around GPT-5.6 Sol: OpenAI says the Cerebras-powered mode can generate up to 750 output tokens per second, or as much as 14 times the speed of Standard processing. The company is testing it with an initial group of customers across coding, commerce, financial research, support, and other interactive applications. 1

That matters when the user is still in the loop. A support agent can resolve a multi-step issue while the conversation is live. An incident-response tool can read logs and recent changes before the outage is over. A research workflow can run another experiment during the workday instead of waiting for an overnight batch. Those are OpenAI's examples, and they point to the actual test: measure completed work per minute, not just tokens per second. 1

The access condition is the product detail to keep. This is a limited preview for selected customers, with access expanding as capacity grows. There is no public price in the announcement. A builder can sign up for updates, but cannot yet budget Ultrafast as a generally available API tier. 1

Google is pushing capability per dollar

Gemini 3.7 Flash is aimed at coding, web development, knowledge work, and agents. Google reports higher scores than Gemini 3.6 Flash on several of its cited evaluations, including 43.6% versus 34.4% on FrontierCode 1.1 Main, 65.3% versus 49.0% on DeepSWE v1.1, and 30.4% versus 17.0% on AutomationBench. Those are Google's reported comparisons, so treat them as a starting point for a task-specific test rather than a universal ranking. 2

The API surface is broad enough for real agent experiments: text, image, video, audio, and PDF inputs; a 1,048,576-token input limit; caching; code execution; file search; function calling; Google Search and Maps grounding; structured outputs; and computer use in preview. Thinking supports low, medium, and high effort. 3

Google's Gemini 3.7 Flash performance-to-cost comparison on the DeepSWE V1.1 evaluation — *Google's chart places Gemini 3.7 Flash against other models by DeepSWE V1.1 score and average cost per task; the comparison is Google's published evaluation, not an independent benchmark. 2*

The price is the sharper change for a builder. Gemini 3.7 Flash is listed at $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026. From Jan. 1, 2027, Google says those prices become $1.50 and $7.50. That makes the deadline part of any cost comparison: a prototype built on the introductory rate needs a second budget before it becomes a production commitment. 2

Developers can try the model in Google AI Studio and Android Studio, while enterprises can use it in Google's agent platform and Gemini Enterprise. For individuals, the consumer path runs through Gemini Spark for Google AI Pro and Ultra subscribers in supported countries. 2

Amazon is moving the agent into the document

Amazon Quick's Microsoft 365 release is about where work happens. The four extensions bring Quick into Word, Excel, PowerPoint, and Outlook on both desktop and web versions of Microsoft 365. Quick can use connected data from Quick Sight, Salesforce, Jira, Slack, SharePoint, and other configured sources while the user stays in the document or inbox. 4

Amazon Quick's extension catalog showing Word, Outlook, Excel, and PowerPoint add-ins — *The AWS console view shows the four Microsoft 365 add-ins together; the practical change is that Quick's connected data and actions can stay inside the tools where the work is already happening. 4*

In Word, Quick can edit text and show a before-and-after comparison. In Excel, it can analyze a sheet, explain formulas, pull data from connected sources, and create charts. In Outlook, it can draft replies, summarize threads, flag messages, and manage recipients. In PowerPoint, it can generate slides that follow the organization's templates and draw on connected data. 4

The prerequisites are more concrete than the phrase "inside Microsoft 365" suggests. Existing Plus, Professional, and Enterprise customers get the extensions without an additional license, but they still need an active Quick application. Admins can deploy the add-ins through the Microsoft 365 admin center or users can install them from the Microsoft add-in store. Outlook commonly needs administrator approval for full Graph API access. Availability is limited to seven AWS Regions, with data staying in the selected Region. 4

What to do with this brief

If latency is your constraint: apply for OpenAI Ultrafast only if your product depends on live interaction. Test the full workflow, including tool calls and retries, because the announcement gives a generation-speed figure rather than a completed-task cost.
If model cost and coding quality are your constraint: run the same agent task on Gemini 3.7 Flash and your current model. Track first-pass success, correction time, token use, and the price change after Dec. 31.
If work is trapped between data systems and Office: check whether your organization already has an eligible Amazon Quick plan, then verify Region and Outlook admin requirements before planning a rollout.

The useful comparison is simple: OpenAI is reducing the wait, Google is lowering the price of a stronger workhorse, and Amazon is reducing the distance between the agent and the document. Choose the test that matches the bottleneck you actually have.