ChatGPT Pro Max is the name of a new, higher ChatGPT subscription that OpenAI appears to be preparing at $500 a month. The plan surfaced on Thursday 24 September 2026 in the code of the ChatGPT web app. Within hours it was confirmed a second way, in a commit to OpenAI’s own open-source Codex repository that teaches the Codex tools to recognise a promax plan. OpenAI has not announced it.
The same week, Google took its voice AI in the other direction. It released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on 15 September, then made Gemini 3.8 Live with Live Avatar generally available to businesses on 24 September.
The two stories pull in opposite directions. OpenAI is testing how much individual professionals will pay for speed and headroom, at a price 25 times that of ChatGPT Plus. Google is selling real-time voice and video agents by the token, at list prices that make an hour of conversation cost well under a dollar in audio.
This article covers both. It sets the ChatGPT Pro Max leak against the primary evidence: the code strings, the Codex commit and OpenAI’s own pricing documents. It also covers what Gemini 3.8 Live does, what it costs and where the limits are. We first reported the Gemini 3.8 Live model names when they appeared on a Google Cloud quota page on 15 September, hours before the launch.
Table of contents
- What the ChatGPT Pro Max Leak Shows
- Pro, Pro (More) and Pro (Max): The New ChatGPT Pro Max Ladder
- What $500 Buys: Speed and the Cerebras Question for ChatGPT Pro Max
- How Much Usage Could ChatGPT Pro Max Include?
- ChatGPT Pro Max Against Claude, Cursor and Google
- Why OpenAI Wants ChatGPT Pro Max Now
- Google Releases Gemini 3.8 Live
- Live Avatar Gives Gemini 3.8 Live a Face
- What Gemini 3.8 Live Costs
- ChatGPT Pro Max and Gemini 3.8 Live: Two Pricing Philosophies
- What Businesses Should Do Now
- Frequently Asked Questions About ChatGPT Pro Max and Gemini 3.8 Live
- References and Further Reading
What the ChatGPT Pro Max Leak Shows
The leak arrived in three pieces over less than four hours. Each piece came from a different source, and together they leave little doubt that a ChatGPT Pro Max plan is being built. What they do not settle is what it will include.
The code string in the ChatGPT web app
At 20:38 UTC on 24 September, Tibor Blaho, who tracks unreleased AI features, posted on X that “OpenAI is preparing a new higher-tier ChatGPT Pro plan.” His screenshot shows a browser developer-tools search for “promax” across the JavaScript that chatgpt.com serves. It returns matches in two bundle files.
The strings are specific. One line defines PROMAX as chatgptpromax next to PRO as chatgptpro. Another defines PROMAX as promax next to prolite and pro. A third orders the plans, returning 5 for PRO and 6 for PROMAX. By the next morning the post had drawn almost 400,000 views.
The $500 price and the “Fastest” promise
The price came from TestingCatalog. Its article, published at 23:56 UTC, says the ChatGPT Pro Max plan “will cost $500 per month”. It also says the plan’s description differs from the existing Pro tier mainly through access to the “Fastest Work and Codex”. Higher usage limits “may also be part of the package, although that is not confirmed yet,” it adds. A short video attached to the article shows the price as $600, which TestingCatalog explains is because the figure includes VAT.
The $500 figure was already circulating a little earlier. Crypto Briefing published it at 21:48 UTC under the headline “OpenAI sets ChatGPT Pro Max subscription at $500”. Neither report cites OpenAI, and OpenAI has not commented.
The Codex commit three and a half hours later
The strongest evidence came from OpenAI itself. At 00:10 UTC on 25 September, a commit by Eric Traut was added to the public openai/codex repository titled “Add Pro Max plan support and update Pro display names”. It was merged at 00:20 UTC as pull request 47971, about three and a half hours after Blaho’s post.
The commit is small, at 51 lines added and 17 removed across 24 files. It does three things. It makes the Codex tools “recognize promax in authentication, account responses, and backend rate limits”. It renames the existing Pro plans on screen. And it includes Pro Max in “experimental context eligibility, paid-plan tooltips, and usage-limit guidance for purchasing credits.” Adding a plan to login and rate-limit code is a step towards supporting real accounts, which makes this the strongest evidence so far that ChatGPT Pro Max is coming.
| Time (UTC) | Source | What it showed |
|---|---|---|
| 24 Sep, 20:38 | Tibor Blaho on X | promax and chatgptpromax strings in the ChatGPT web app, ranked above Pro |
| 24 Sep, 21:48 | Crypto Briefing | $500 a month price |
| 24 Sep, 23:56 | TestingCatalog | $500 price, “Fastest Work and Codex”, $600 including VAT in a video |
| 25 Sep, 00:10 | openai/codex commit | promax plan added to login, account and rate-limit code |
| 25 Sep, 00:20 | Pull request 47971 merged | Display names changed to Pro, Pro (More) and Pro (Max) |
Pro, Pro (More) and Pro (Max): The New ChatGPT Pro Max Ladder
The most revealing part of the commit is the renaming. OpenAI currently sells ChatGPT Pro in two sizes, a $100 plan with five times the usage of Plus and a $200 plan with 20 times the usage of Plus. Inside the code they are called prolite and pro. Until this commit, the Codex terminal interface displayed them as “Pro Lite” and “Pro”.
After the commit, the Codex tools show three names. prolite becomes “Pro”, pro becomes “Pro (More)” and promax becomes “Pro (Max)”. A new ChatGPT Pro Max tier on top pushes the whole ladder up a rung.
| Plan code | Old Codex label | New Codex label | Price a month | Usage vs Plus |
|---|---|---|---|---|
prolite | Pro Lite | Pro | $100 | 5x |
pro | Pro | Pro (More) | $200 | 20x |
promax | Not present | Pro (Max) | $500 (reported) | Not published |
Pro Lite becomes plain Pro
The label change matters more than it looks. On OpenAI’s own Codex pricing page, Pro is already sold as one card priced “From $100” with a choice of “5x or 20x higher rate limits than Plus”. Calling the $100 plan “Pro” in the product, rather than “Pro Lite”, matches that framing. It makes the entry plan the default Pro and turns the others into upgrades. With ChatGPT Pro Max above it, Pro becomes a three-step ladder rather than two sizes of one plan.
What Pro Max unlocks in Codex today
The code also shows what OpenAI treats a ChatGPT Pro Max account as. It qualifies for “experimental context”, a feature the Codex code limits to Plus, Pro, Pro Lite and now Pro Max accounts signed in with ChatGPT. It gets the paid-plan tips. And when a Pro Max user hits a limit, Codex shows the same message it shows Pro users: “You’ve hit your usage limit. Visit https://chatgpt.com/codex/settings/usage to purchase more credits”.
That last detail matters. Even at $500 a month, the code assumes ChatGPT Pro Max users can run out and will be pointed to buy credits. Nothing in the commit suggests unlimited use.
What $500 Buys: Speed and the Cerebras Question for ChatGPT Pro Max
TestingCatalog’s single line of plan description, “Fastest Work and Codex”, is the main clue to the pitch. It suggests ChatGPT Pro Max is built for developers, researchers and professionals who run long agentic workloads, where speed matters more than it does in a chat.
Work and Codex share one allowance
OpenAI’s documentation says that “ChatGPT Work and Codex share usage. Both use the same pricing, credits, and usage limits.” Work is designed for longer tasks, and Codex can run coding jobs for hours. A plan that makes both of them faster makes every long task finish sooner, which is worth more to someone paying for an agent’s time than to someone asking quick questions.
Fast mode already costs 2.5 times the credits
OpenAI already sells speed, and it prices it in credits. According to OpenAI’s Speed documentation, Fast mode for GPT-6 Astra, GPT-6 Sol and GPT-6 Luna “consumes credits at 2.5x the Standard rate where available.” For older GPT-5.6, GPT-5.5 and GPT-5.4 models, Fast mode increases model speed by 1.5 times. Users switch it on with /fast on in the Codex CLI.
So speed on today’s plans burns through a user’s allowance much faster. The obvious product for ChatGPT Pro Max is a plan where the fastest mode is the default, or is far cheaper in credits, and where limits are high enough to use it all day. That is our reading of “Fastest Work and Codex”, not something OpenAI has said.
Ultrafast and Cerebras
TestingCatalog also points to Cerebras. On 13 August, according to TestingCatalog, OpenAI previewed an Ultrafast mode for GPT-5.6 Sol that could make the model up to 14 times faster. It was available to “a limited set of customers”, with others able to sign up for updates. TestingCatalog says OpenAI already uses Cerebras for that latency-focused path, and that the “Fastest” tier could rely on more Cerebras capacity. It adds that OpenAI “has not confirmed a direct connection between Cerebras and Pro Max.” For background on the chipmaker, see our look at Cerebras’s IPO filing.
How Much Usage Could ChatGPT Pro Max Include?
OpenAI publishes usage estimates for its current plans on its Codex pricing page, in messages per five-hour window. It has published nothing for Pro Max. But the existing ladder gives a way to frame the possibilities.
| Model (messages per 5 hours) | Plus, $20 | Pro 5x, $100 | Pro 20x, $200 |
|---|---|---|---|
| GPT-6 Astra | 5–45 | 25–225 | 100–900 |
| GPT-6 Sol | 15–150 | 70–700 | 300–3,000 |
| GPT-6 Luna | 350–3,000 | 1,750–14,000 | 7,000–56,000 |
The figures come from OpenAI’s Codex pricing page as it stood on 25 September. OpenAI stresses that “these estimates are not fixed message limits” and that “Weekly limits may also apply.”
Two ways to price the top rung
On the current ladder, Pro 5x gives five times Plus usage for five times the price. Pro 20x gives 20 times the usage for ten times the price. The $200 plan is therefore twice as generous per dollar as the $100 plan.
If ChatGPT Pro Max kept the $100 plan’s ratio, $500 would buy 25 times Plus usage. If it kept the $200 plan’s ratio, it would buy 50 times. Those are the two simplest anchors, and they are our arithmetic, not OpenAI’s numbers. The TestingCatalog report suggests the difference may be speed rather than volume, in which case the multiple could be lower still.
The VAT line for UK buyers
TestingCatalog’s video showed $600, and says the difference is VAT. That is exactly 20% on top of $500, the standard UK rate. UK businesses that can reclaim VAT will care about the net figure, but individuals paying personally should budget for the gross one. At $600 a month, ChatGPT Pro Max would cost $7,200 a year per seat including tax.
ChatGPT Pro Max Against Claude, Cursor and Google
At $500, ChatGPT Pro Max would sit well above the top individual plans from OpenAI’s main rivals. Most of those top plans cluster around $200 to $250.
| Plan | Top individual price a month | What the vendor sells it on |
|---|---|---|
| ChatGPT Pro Max (reported) | $500 | “Fastest Work and Codex” |
| Google AI Ultra | About $250 in the US | Google’s top consumer AI plan |
| Claude Max | From $100 for 5x; $200 for 20x | 5x or 20x the usage of Claude Pro |
| Cursor Ultra | $200 | For “agent power users” |
| ChatGPT Pro (20x) | $200 | 20x Plus usage; new sign-ups paused on 10 September |
Prices are as listed on each vendor’s pricing page on 25 September, except Google AI Ultra and the $200 Claude Max option, which are TestingCatalog’s figures. Anthropic’s page describes Claude Max as “From $100” with a choice of “5x or 20x more usage than Pro”.
Why a gap this large is not unusual
The jump from $200 to $500 looks steep next to rivals, but it is a common pricing move. When one plan strains capacity, a vendor can add a more expensive plan above it. The top plan then carries the users who are willing to pay most, which protects capacity for everyone else. ChatGPT Pro Max would do exactly that for OpenAI’s heaviest users.
Why OpenAI Wants ChatGPT Pro Max Now
The timing is not an accident. Two things happened in September that make a higher tier useful to OpenAI.
The $200 pause
On 10 September, OpenAI paused new subscriptions to its $200 Pro plan. OpenAI’s Thibault Sottiaux said the $200 plan was under the most strain from demand for GPT-6 Astra, and called the pause “the smallest step” OpenAI could take. All other plans and the API stayed available. We covered the decision and what it means for existing subscribers in OpenAI Puts Pro Subscriptions on Hold Due to Astra Demand. TestingCatalog’s 25 September report still describes the $200 tier as paused.
A ChatGPT Pro Max plan gives OpenAI another way to manage that demand. Heavy users who want more than the $200 plan can offer could move to a tier priced to cover the extra compute. That, in turn, could ease the pressure that led to the pause.
DevDay on 29 September
OpenAI’s developer conference, DevDay 2026, takes place in San Francisco on Tuesday 29 September. TestingCatalog expects OpenAI to discuss APIs, developer tooling and new subscription tiers for its platform there. A plan built around the fastest Work and Codex would fit that audience. We have previewed what else is expected in our DevDay coverage of OpenAI’s managed agents and environments.
What is still unconfirmed
OpenAI has not announced ChatGPT Pro Max, its price, its limits or a launch date. The $500 figure comes from reporting, not from OpenAI. The Codex commit proves the plan exists in OpenAI’s systems, but it does not reveal what the plan includes. Plans that appear in code sometimes launch under different names or prices, and some never launch at all.
Google Releases Gemini 3.8 Live
While OpenAI’s plans leaked, Google shipped. On 15 September it launched two new live dialogue models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. The Gemini API changelog lists both as generally available that day, under the model names gemini-3.8-live and gemini-3.8-live-extended-thinking.
Two models, two jobs
Google describes Gemini 3.8 Live as “built for scale and cost efficiency, combining conversational intelligence with fluid dialogue and visual grounding.” Gemini 3.8 Live Extended Thinking is “built for high-complexity tasks, with increased intelligence and multi-step reasoning.”
The Extended Thinking model “reasons and speaks simultaneously.” It uses early verbal cues such as “Let me check that…” to acknowledge a request, then narrates its progress while it works through background tasks. The idea is to avoid the long silence a voice agent normally leaves while it thinks.
The benchmark claims
Google made several performance claims at launch. They are Google’s own figures, and the table lists them as Google reported them.
| Benchmark | Model | Result Google reported |
|---|---|---|
| Artificial Analysis Speech to Speech Quality Index | 3.8 Live Extended Thinking | 82.6, first place overall |
| Ï„-Voice agentic task completion | 3.8 Live Extended Thinking | 68.6% |
| Sierra Ï„-Voice-banking | 3.8 Live Extended Thinking | 35.1% |
| Big Bench Audio | 3.8 Live Extended Thinking | 97.7% |
| Speech Agent Arena | 3.8 Live | Second place |
| ServiceNow EVA-Bench | Both | On the accuracy and quality Pareto frontier |
The banking figure is worth pausing on. A model that leads its benchmark still completes only about a third of the tasks in a realistic banking scenario. Voice agents are improving quickly, but they are not yet at the point where businesses can hand them complex customer requests without supervision.
97 languages and background tools
Gemini 3.8 Live processes visual input in near real time, so it can respond to what a user shows it on a camera or shared screen. That is computer vision built into the same model that listens and speaks, rather than a separate system bolted on. It detects and switches between 97 supported languages mid-conversation. And it runs tools and API calls in the background while the conversation continues, so it can acknowledge a request and keep talking while the task finishes.
That last feature, asynchronous tool calling, is the practical one for business use. A customer service agent that goes quiet for five seconds while it looks up an order feels broken. One that says “I’m pulling that up now” and keeps talking does not.
Where you can use it
| Audience | Gemini 3.8 Live | Gemini 3.8 Live Extended Thinking |
|---|---|---|
| Developers | Gemini API and Google AI Studio | Gemini API and Google AI Studio |
| Enterprises | Private preview in Gemini Enterprise; Gemini Enterprise for Customer Experience coming soon | Private preview in Gemini Enterprise; Workspace business customers coming soon |
| Consumers | Search Live | Gemini Live; Docs for Google AI Pro and Ultra subscribers; Gmail and Keep for all Google AI subscribers |
Google named several developer platforms that support the Gemini Live API, including Agora, LiveKit, Pipecat, LangChain and Vercel. It also said Salesforce, Genspark and Lumeris were working with the new models. Gemini’s desktop app is picking up the same live capabilities, as we reported when Gemini’s desktop Task mode added Gemini Live and computer use settings.
Live Avatar Gives Gemini 3.8 Live a Face
Nine days after the launch, on 24 September, Google introduced Gemini 3.8 Live with Live Avatar. It pairs the live dialogue model with near real-time video generation, so a voice agent can appear on screen as an animated character that listens, sees and speaks. Google Cloud announced it as generally available in Gemini Enterprise at the same time.
What Live Avatar does
Google says the feature delivers “precise lip-syncing, natural expressions, and fluid turn-taking.” It keeps the asynchronous tool calling of the base model, so an avatar can fetch data in the background while the conversation continues. Google’s demonstration shows an avatar checking in a hotel guest while tools run in the background.
The avatar also inherits the 97 languages. Google says it “can seamlessly transition across 97 languages without degrading video fidelity or introducing visual drift,” adapting lip-sync and expressions as it switches. Google Cloud says the feature works across web, mobile and interactive kiosks.
Enterprise only, with custom avatars on an allowlist
Live Avatar is a business product. It is available in Gemini Enterprise, with US and EU endpoints, provisioned throughput and enterprise data governance. Businesses can choose from a library of preset avatars. Custom avatars, built from a single high-quality reference photo and an audio sample, are “currently available only through enterprise allowlisting.”
Google says that restriction is there to “safeguard identity and prevent misuse.” All generated audio and video carries SynthID, Google’s imperceptible watermark, so the output can be identified as AI-generated. Google Cloud also noted that Gemini 3.8 Live Extended Thinking “remains in private preview” on its enterprise platform, even though developers can already use it through the Gemini API.
Limits in the model card
The Gemini 3.8 Audio model card, published on 15 September, is more candid than the launch posts. It says the Live models accept audio, images, video and text with a context window of up to 128K tokens, and output up to 64K tokens. With Live Avatar switched on, output drops to 24K tokens.
It also says Live Avatar “can support a few minutes of continuous interaction, rather than extended hours.” That is a real limit for customer service, where calls can run long. The card gives the models’ knowledge cutoff as January 2025, and says they are based on Gemini 3 Pro.
Early customers
Google Cloud named four early users. Cox Automotive built a shopping assistant for Autotrader that highlights parts of the screen and uses tools to guide car buyers through search, comparison and financing. Salesforce is combining Gemini 3.8 Live with its Agentforce platform. Specs praised improvements in voice activity detection and latency.
The most striking quote came from Akhilesh Damaraju, chief executive of Equal AI, who said its assistant “handles over a million live calls daily across nine Indian languages.” According to Damaraju, Gemini 3.8 Live improved interruption handling, multilingual conversations and tool-call reliability.
What Gemini 3.8 Live Costs
Google prices Gemini 3.8 Live by the token, and it publishes a per-minute conversion for audio. The table combines the Gemini API price list with the Google Cloud price list, which adds the rate for avatar video.
| Gemini 3.8 Live | Price per 1 million tokens | Per-minute equivalent |
|---|---|---|
| Text input | $0.75 | – |
| Audio input | $3.00 | $0.005 |
| Image and video input | $1.00 | $0.002 |
| Text output | $4.50 | – |
| Audio output | $12.00 | $0.018 |
| Video output (avatar) | $1.00 | Not published |
Those rates match the list prices of Google’s earlier Gemini 3.1 Flash Live preview, so the upgrade comes without a price rise. The Gemini API also offers a free tier, and Grounding with Google Search is supported, with 5,000 free search requests a month.
What an hour of conversation costs
The per-minute figures make a rough estimate easy. Take a one-hour call in which the user’s audio streams for the full 60 minutes and the model speaks for half of it. Audio input costs 60 × $0.005, or $0.30. Audio output costs 30 × $0.018, or $0.54. That comes to $0.84 an hour before text, tools, video or search.
For comparison, OpenAI prices its GPT-Live-1 voice API at $0.05 a minute, billed per second including silence, as we reported when OpenAI launched GPT-Live-1 and its Agents API. An hour at that rate is 60 × $0.05, or $3.00. The two pricing models measure different things, so treat the comparison as a guide rather than a like-for-like quote.
ChatGPT Pro Max and Gemini 3.8 Live: Two Pricing Philosophies
Put side by side, the two stories show the two ways AI companies now charge for their most capable products.
ChatGPT Pro Max is a flat subscription for a person. It sells access, speed and headroom for a fixed monthly fee, whatever the user does with it. Its customer is a professional who spends the working day inside ChatGPT Work or Codex and whose time is worth more than $500 a month.
Gemini 3.8 Live is metered infrastructure for a business. It sells minutes of conversation to companies building their own agents, and the bill scales with every customer call. Its customer is a contact centre, a retailer or a bank that wants to put a voice or a face on its service.
Where the two meet
The two worlds overlap in the middle. A developer building a voice agent on Gemini 3.8 Live might well write the code with Codex on a ChatGPT Pro Max plan. And both companies are racing to build AI agents that talk, look and act in real time. OpenAI’s GPT-Live-1 competes directly with Gemini 3.8 Live for developers, while Google AI Ultra competes with ChatGPT Pro Max and ChatGPT Pro for individual power users.
A useful yardstick
The numbers give a sense of scale. $500 is the reported monthly price of one ChatGPT Pro Max seat. At the $0.84-an-hour estimate above, the same $500 would pay for roughly 595 hours of Gemini 3.8 Live audio. The comparison is loose, because one is a seat and the other is a service. But it shows how differently the two products are priced.
What Businesses Should Do Now
Neither announcement needs immediate action, but both change the options for next year’s budgets.
If you are weighing ChatGPT Pro Max
Wait for OpenAI to publish the limits. The value of ChatGPT Pro Max depends entirely on how much faster and how much larger its allowance is than the $200 plan, and none of that is public. Look at your current usage first. If your heaviest users rarely hit their Pro limits, a $500 tier is unlikely to pay for itself. If they hit limits every week and lose time waiting, it might. For most teams, the right comparison is against API usage or ChatGPT Business, not against Plus.
If you are building a voice agent
Gemini 3.8 Live is now a credible option at a low list price, and Google offers EU endpoints for it in Gemini Enterprise for businesses with data residency needs. Test it against your real calls, not against benchmarks. The 35.1% banking score is a reminder that complex requests still fail often. Plan for a human handover, and check the few-minute limit on Live Avatar before designing a long video session.
If you are planning an AI programme
Both products show how fast pricing and packaging are moving. Plans appear in code before they are announced, and model versions change every few weeks. Our AI models, tools and releases hub tracks the changes as they land, and our team can help you build a plan that survives them.
Frequently Asked Questions About ChatGPT Pro Max and Gemini 3.8 Live
What is ChatGPT Pro Max?
It is a new ChatGPT subscription tier above the existing $100 and $200 Pro plans. It appeared in the ChatGPT web app’s code on 24 September 2026 and in a commit to OpenAI’s Codex repository on 25 September. OpenAI has not announced it.
How much will ChatGPT Pro Max cost?
TestingCatalog reported $500 a month, or $600 including VAT. OpenAI has not confirmed a price.
What will ChatGPT Pro Max include?
The reported plan description promises the “Fastest Work and Codex”. Higher usage limits may also be included, but that is unconfirmed. The Codex code treats Pro Max like other Pro plans, including the option to buy credits when a limit is reached.
When will ChatGPT Pro Max launch?
There is no date. OpenAI’s DevDay conference on 29 September is a possible venue for an announcement, but nothing confirms it.
What is Gemini 3.8 Live?
It is Google’s newest live dialogue model for real-time voice conversations, launched on 15 September 2026 alongside Gemini 3.8 Live Extended Thinking. It supports 97 languages, near real-time visual input and background tool calls.
What is Live Avatar?
It is a Gemini 3.8 Live feature that adds an animated, lip-synced video character to a voice agent. It became generally available in Gemini Enterprise on 24 September 2026. Custom avatars need enterprise allowlisting.
How much does Gemini 3.8 Live cost?
List prices are $3.00 per million audio input tokens and $12.00 per million audio output tokens, or about $0.005 and $0.018 a minute. Text, image, video and avatar video are priced separately.
References and Further Reading
OpenAI is preparing a new higher-tier ChatGPT Pro plan (Tibor Blaho on X)
OpenAI prepares new $500/month Pro Max plan for ChatGPT (TestingCatalog)
Add Pro Max plan support and update Pro display names (openai/codex on GitHub)
Pull request 47971 (openai/codex on GitHub)
OpenAI sets ChatGPT Pro Max subscription at $500 (Crypto Briefing)
Speed (OpenAI Codex documentation)
OpenAI announced Ultrafast mode for GPT-5.6 Sol (TestingCatalog on X)
Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking (Google)
Introducing Gemini 3.8 Live with Live Avatar (Google)
Power your agents: Gemini 3.8 Live with Live Avatar is now generally available (Google Cloud)
Gemini 3.8 Audio model card (Google DeepMind)
Gemini API release notes (Google AI for Developers)
Gemini Developer API pricing (Google AI for Developers)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
