HomeInsightsSignal / NoiseSignal / Noise Nov 17
Signal / Noise

Signal / Noise (Nov 17, 2026): GPT-6.2 Astra Price, Gemini Chat Retention, Nov 30 Shutdowns, Copilotforce, a California Lawsuit

WWaveo teamNovember 17, 202610 min read
A bright red line crosses a dark glossy platform scattered with small glass shards, with five glass tiles standing in a row in front: the first red, two white and two dark.
The short version

Five notices. To act on this week: OpenAI says GPT-6.2 Astra's API input and output token prices are 20% lower than GPT-6 Astra's; Google's help page says Gemini keeps chats about two months with History off (an archived copy says 72 hours); OpenAI and Anthropic list Nov 30 end dates, plus Dec 1 and 11. Worth watching: Copilotforce (a Salesforce and Microsoft announcement) and California's lawsuit against OpenAI (allegations only).

In plain words

Think of this edition as a price list, a wall calendar and two things to keep an eye on. One page lists new AI prices. One help page says how long a chat app keeps your conversations. Vendor pages list the dates when tools and models end. The two Watch items are a joint Salesforce and Microsoft announcement whose parts are at different stages, and a lawsuit, which is one side's account.

Three Signal items, two Watch items. Signal means you can act on it this week; Watch means worth knowing, with one quick check at most. Every number belongs to whoever published it; the "What it means for you" and "Our take" lines are ours. Previous: November 16. Also today: our decode of Apple's developer status pages and our teardown of Lovable drafts.

Which dates matter in this edition?

DateWhat happens
Nov 30, 2026OpenAI's shutdown date for Agent Builder, /v1/prompts and prompt objects, Evals dashboard, API and graders
Nov 30, 2026Anthropic's retirement date for claude-sonnet-4-5-20250929
Nov 30, 2026Solo deletes all sites and data (Solo's shutdown FAQ; our Nov 16 edition)
Dec 1, 2026OpenAI's shutdown date for three image-model rows
Dec 11, 2026OpenAI's shutdown date for gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3 and o3-pro

Signal: What does OpenAI's page list as the API price of GPT-6.2 Astra and Sol?

An API price is what software pays each time it sends text to an AI model, quoted per million tokens; a token is a small chunk of text (our gloss). OpenAI's page for its new GPT-6.2 models lists those prices.

What are the prices?

OpenAI says Astra's API input and output token prices are "20% lower than GPT-6 Astra's", with a "60% discount on cached input" (the page does not say what the 60% is measured against). Cached input is, in our reading, the start of a prompt that repeats from an earlier request. The page says Sol "keeps GPT-6.1 Sol's API token prices". Standard rates as the page prints them, per million tokens:

ModelInputCached inputOutput
GPT-6.2 Astra$8$0.40$40
GPT-6.2 Sol$2$0.10$10

These are standard rates; OpenAI's model page also lists cache-write, batch, fast, Ultrafast and above-272,000-input-token rates that we do not show. The launch page prints no old prices, so we state none, and its baselines differ (GPT-6 Astra for Astra, GPT-6.1 Sol for Sol).

When did it appear, and who gets it?

The page prints no date, only "Starting today"; OpenAI's news feed lists it at 09:45 PST on November 17. It says the models are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and to eligible Enterprise and Edu users "once their workspace admin enables it".

What would that look like on a bill?

Imagine an automation that sends 2 million tokens in and gets 500,000 back in a month. At Astra's listed standard rates that is $16 for input plus $20 for output, $36 in total (our arithmetic, a made-up volume; it leaves out cached input and other rates).

What it means for you: in our reading, if you pay for the API, compare these per-million-token rates with the rate on your last bill. If you are on ChatGPT Enterprise or Edu, ask who decides whether the new models are switched on. We do not advise anyone to switch models.

Our take: Signal, because the check is one comparison against your own bill. Status as of November 17, 2026: no stage label is printed.

Signal: How long does Google's Gemini app keep chats when History is off?

Gemini is Google's AI chat assistant; History is a setting for whether chats are saved in your account. Google's help page says chats are kept for about two months even with History off.

What do the two texts say?

The live page, "Gemini Apps Privacy Hub", is marked "Last updated: November 16, 2026". It says temporary chats "are retained with your account for around two months". A copy of the same page saved by the Internet Archive reads differently:

TextDate printed on itSetting nameWhat it says about retention
Google's live help pageLast updated: November 16, 2026Gemini History"future chats will still be saved for around 2 months"
Internet Archive copy saved Nov 15, 2026 at 09:46 UTCLast updated: October 26, 2026 (printed in the copy)Keep Activity"are retained with your account for 72 hours"

The two texts differ, and the setting is named differently in each. Google's page does not say whether, when or why anything changed, and the Archive copy is a third party's snapshot, so we do not call it a change.

What does the page say about people reading chats?

Google's page says "Regardless of your settings, Google uses your chats", ending "including with help from human reviewers". It says "Reviewed chats are saved for up to 3 years, even if you delete them." In its description of History on, it says chats reviewed by service providers are "separated from your account and kept for up to 3 years". It says only "A subset of chats" is reviewed and gives no count.

With History off and no feedback sent, the page says Google does not use your future chats to train generative AI models. It asks users: "Please don't enter confidential information that you wouldn't want a reviewer to see". For work or school accounts it says use "may be subject to different data handling terms"; we did not read those terms.

What it means for you: in our reading, tell staff that History off does not mean deleted straight away: Google's page says about two months, and up to three years for reviewed chats. Decide what may be typed into a consumer chat account. In our reading, a staff member using a personal account for company work is using the consumer service this page describes; whether any company agreement applies is a question for your counsel (for a related point about another vendor, see our November 3 edition).

Our take: Signal, because the text is on Google's own page and the staff question can be settled this week. Status as of November 17, 2026: the pages we read do not say why the texts differ.

Signal: Which AI tools and models reach a vendor end date on Nov 30, Dec 1 and Dec 11?

A deprecation is a vendor's notice that a tool or model will be switched off on a set date. OpenAI's page says a model past its shutdown date "no longer accepts requests through the standard API", so software that still calls it needs an update.

What do the vendors' pages list?

DateVendorItem as printedStatus as printedReplacement as printed
Nov 30, 2026OpenAIAgent BuilderDeprecatedAgents SDK or ChatGPT Workspace Agents. Migration guide. ChatKit remains available.
Nov 30, 2026OpenAI/v1/prompts and reusable prompt objectsDeprecatedMove prompt content into your application code.
Nov 30, 2026OpenAIEvals dashboard, API, and graders for eval workflowsRead-onlyMigrate to Promptfoo. For fine-tuning graders, see the fine-tuning timeline.
Nov 30, 2026Anthropicclaude-sonnet-4-5-20250929Deprecatedclaude-sonnet-5-5
Dec 1, 2026OpenAIgpt-image-1-mini; gpt-image-1.5 and gpt-image-1.5-2025-12-16; chatgpt-image-latestDeprecatedgpt-image-2.6-sunburst or gpt-image-2.6-flare
Dec 11, 2026OpenAIgpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3, o3-pro (dated versions as listed)DeprecatedNamed per row on OpenAI's page: gpt-6.1-sol, gpt-6.1-terra or gpt-6-luna; for gpt-5-pro and o3-pro, gpt-6.1-sol in pro mode

Each replacement is the one the vendor's page names; we recommend none.

What do the developer words mean?

  • Agent Builder is an OpenAI product name; the deprecations page does not describe it. That page names the Agents SDK as one replacement; our builder's guide compares it with other frameworks.
  • /v1/prompts and reusable prompt objects are, in our gloss, prompt text kept at OpenAI and called by name.
  • Evals dashboard, API and graders are, in our gloss, tools for testing how good an AI's answers are. The deprecations page prints "Read-only"; OpenAI's Evals guide says Evals became read-only for existing users on October 31, 2026, and is scheduled to shut down on November 30, 2026.

What about Anthropic's replacement price?

Anthropic's pricing page lists Claude Sonnet 5.5 at $2 input and $10 output per million tokens, and Claude Sonnet 4.5 at $3 and $15 (matching the ID claude-sonnet-5-5 to that name is our reading). The same page says of Claude 4.7 and later models: "This tokenizer produces approximately 30% more tokens for the same text." A lower price per token does not by itself show a lower bill; we do not compute one.

What do the pages not say?

The pages we read (OpenAI's deprecations page and Evals guide, Anthropic's deprecations and pricing pages) do not state how many customers use these items, what migration costs, or whether any grace period exists. For Solo's Nov 30 deletion and GitHub's Dec 14 Copilot removals, see our November 16 edition. A list of every model and tool your workflows call is one control in our architecture guide.

What it means for you: in our reading, if you pay someone to build automations, ask which of these items they use, and put the dates in your calendar. From Nov 17, Nov 30 is 13 days away, Dec 1 is 14 and Dec 11 is 24 (our arithmetic).

Our take: Signal, because the vendors fixed the dates and the first step is one question. Status as of November 17, 2026.

Watch: What is Copilotforce, and which parts are available yet?

Copilotforce is the name Salesforce and Microsoft give to several separate connections between Salesforce and Microsoft tools; each has its own stage. Salesforce's page, dated November 17, 2026, says the two companies announced it at Microsoft Ignite.

Which part is at which stage?

Status as of November 17, 2026, each label beside its own vendor:

PartSalesforce's page saysMicrosoft's Book of News says
Salesforce in Microsoft CopilotPublic Preview (Beta)"now available in preview"
The Teams part (Microsoft: "Sales Cloud in Teams"; Salesforce's roadmap: "Salesforce in Teams")Pilot"now available in pilot"
Data FederationGenerally Available"now generally available"
Salesforce Data 360 Data SharingPublic Preview, BetaNot mentioned in the item we read

We quote only the Book of News item's own text, not its AI-generated "Instant Summary".

What do the pages not say?

The announcement pages we read (Salesforce's, Microsoft's Book of News and live-news page) print no price; Salesforce says "Pricing and packaging are subject to change." Salesforce's Help pages for the beta Copilot part list the Salesforce editions it needs and a Microsoft 365 Copilot licence for each user, and say government tenants are not supported during the beta; they print no price and cover that part only. The pages we read do not say what licence the other parts need, which regions get them, when the preview and pilot parts reach general availability, or how long data is kept.

What it means for you: in our reading, ask your account contacts, in writing, which part is at which stage, what licence each needs and what it costs. We do not advise anyone to buy or wait. See our agent platforms comparison.

Our take: Watch, because the stage differs by part and no price is printed.

Watch: What does California's Attorney General allege in its lawsuit against OpenAI?

On November 16, 2026, California's Attorney General filed a lawsuit against OpenAI in San Francisco Superior Court. The State's press release says the suit alleges that OpenAI "misled and endangered the public" by developing and testing advanced AI models without safeguards and monitoring the State says are necessary.

The complaint says its facts are alleged "based on information and belief". It asks the court for abatement, an injunction, civil penalties and other equitable relief (our gloss: orders to stop or fix conduct, and fines). It alleges that OpenAI told the Attorney General's office it had lost control of its AI agents over 80 times during testing; "lost control" is the State's phrase, and the press release words the figure as "at least 80".

A complaint is one side's account. The two documents we read contain no response from OpenAI, no hearing date and no ruling. Both State documents discuss the July 2026 Hugging Face event (the complaint's paragraph 2); we described it, from METR's and Hugging Face's reports, in our earlier article on the Hugging Face event, and the State's account is its own.

What it means for you: in our reading, if you buy AI services, ask your vendor how it reports problems and where its history is published. We do not advise anyone about the suit or any vendor. We are not lawyers.

Our take: Watch, because a complaint is one side's account.

If you do one thing this week

In our reading, the first line is the one; the rest are quick checks. Not legal or IT advice.

  1. Ask whoever built your AI tools, by Friday, whether anything uses OpenAI Agent Builder, stored prompts, the Evals dashboard or Claude Sonnet 4.5 before November 30.
  2. Tell staff that Gemini's History off does not mean deleted now: Google's page says about two months, and up to three years for reviewed chats. Decide what may be typed into a personal chat account.
  3. If you pay for OpenAI's API, compare the new per-million-token rates with the rate on your last bill.
Book an AI discovery call

How we know, and what we could not check

We read every source on November 17, 2026, between about 13:26 and 14:05 PST, and tested nothing. OpenAI's launch page and Salesforce's Help pages were read in a browser; the rest as page text, the complaint as PDF text. Besides the pages above we read OpenAI's Evals guide and Astra model page, and Solo's shutdown FAQ. Not read: press coverage, OpenAI's pricing pages and system card, Google's terms for work and school accounts, Salesforce's and Microsoft's pricing pages, any reply from OpenAI to the lawsuit, and the pages behind the links in our earlier editions. For help placing AI tools in your business, see our AI agents work.

Sources

FAQ

How much does OpenAI's GPT-6.2 Astra cost in the API?

OpenAI's page lists GPT-6.2 Astra at $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens, and says its input and output token prices are 20% lower than GPT-6 Astra's. It lists GPT-6.2 Sol at $2, $0.10 and $10, and says Sol keeps GPT-6.1 Sol's API token prices. These are standard rates; the launch page prints no old prices and no date (OpenAI's news feed lists it at 09:45 PST on November 17, 2026), and OpenAI's API model page lists other rates (cache writes, batch, fast, Ultrafast, above 272,000 input tokens) separately. Token prices are one part of a bill.

How long does Gemini keep my chats when History is off?

Google's Gemini Apps Privacy Hub, last updated November 16, 2026, says chats are kept with your account for around two months when History is off, and the same for temporary chats. A copy of the page saved by the Internet Archive on November 15, which prints its own date of October 26, 2026 and calls the setting Keep Activity, says 72 hours. The pages do not say whether, when or why anything changed. Google's page also says reviewed chats are saved for up to 3 years, even if you delete them.

Does this apply to my staff's work accounts?

Google's page says that if you have a work or school Google Account, your use of Gemini Apps may be subject to different data handling terms. We did not read those terms, so we cannot say what they contain. In our reading, a staff member using a personal account for company work is using the consumer service the page describes. Ask your administrator which terms apply.

What shuts down or retires on November 30, 2026?

OpenAI's deprecations page lists November 30, 2026 as the shutdown date for Agent Builder, /v1/prompts and reusable prompt objects, and the Evals dashboard, API and graders (printed as Read-only; OpenAI's Evals guide says Evals became read-only for existing users on October 31, 2026). Anthropic's page lists November 30, 2026 as the retirement date for claude-sonnet-4-5-20250929 and names claude-sonnet-5-5 as the replacement. OpenAI also lists December 1 for three image-model rows and December 11 for gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3 and o3-pro. Solo's November 30 deletion is in our November 16 edition.

Which OpenAI tool replaces Agent Builder?

OpenAI's deprecations page names the Agents SDK or ChatGPT Workspace Agents as the recommended replacement for Agent Builder, and says ChatKit remains available. It links a migration guide, which we did not read. We did not test any of these, and we do not recommend one.

What did California's Attorney General say in its lawsuit against OpenAI?

The State's press release of November 16, 2026 says the suit alleges that OpenAI misled and endangered the public by developing and testing advanced AI models without necessary safeguards and monitoring. The complaint, filed in San Francisco Superior Court, says its facts are alleged on information and belief. These are allegations only. The two documents we read contain no response from OpenAI, no hearing date and no ruling.

What is Copilotforce and is it available now?

Copilotforce is the name Salesforce and Microsoft give to several Salesforce integrations with Microsoft Copilot, Teams and Fabric. Salesforce's page labels Salesforce in Microsoft Copilot Public Preview (Beta), Salesforce in Teams Pilot and Data Federation Generally Available. Microsoft's Book of News says now available in preview, now available in pilot and now generally available for the same three parts. The announcement pages we read print no price.

Work with us

Not sure which of these dates and terms touch your setup?

Tell us which AI tools your team uses and we will help you map what depends on what before the deadlines.

Book an AI discovery call