Google

Google announced that agentic video understanding was generally available. The announcement stated that the capability had reached general availability.

At general availability Google described Argon’s headline number as "an industry-leading 1M unified input-output token limit via the Gemini API", and pointed developers to a new continuation protocol for working within it. The announcement two weeks earlier, recorded on this page, had described the same million as the model’s output-token limit, raised from 64,000. A unified million that input, reasoning and output all draw on is a much tighter envelope than a million of output on top of whatever was sent in, and nothing in either post reconciles the two descriptions.

On the day Argon went generally available Google also published its robotics case for it, and the claim is that no separate robotics model is needed: the same general-purpose model scores 87 percent on the Embodied Reasoning benchmark “using pure reasoning”, which Google says beats every other frontier model. The post walks through robot control, reconstructing calibrated simulations from camera feeds or hand-drawn sketches, and self-improvement, including Argon building a MuJoCo twin of a foot-pedal bin to rehearse before working a Boston Dynamics Spot on the real one. It is Google’s own benchmark figure, not an independent result.

Google made Gemini 4 Argon generally available on 13 October, giving "developers, enterprises, scientists, and consumers" access to the endpoint gemini-4-argon, thirteen days after announcing it as a rollout to a set of trusted cyber defenders. The same post expands that original programme too — "hundreds of additional Fairwind Program partners" — and makes plain that the two are not the same model: cybersecurity defenders can apply to Fairwind for Argon "without cyber guardrails", while the public endpoint keeps them.

Google’s announcement footnotes the end of the introductory rate this page recorded on 30 September: from 1 December 2026, after the introductory period expires on 30 November, Argon costs $4 per million input tokens and $20 per million output tokens — double the launch price. The footnote adds that this "applies to the Argon model intended for public access only". The pricing page shows the same schedule for context caching, $0.10 doubling to $0.20 with implicit caching only, and lists no free tier for Argon at all, where Gemini 3.8 Flash is still free of charge.

Google Cloud said its TPU 8i system delivers 80% better price-performance than the prior generation. Google Cloud linked the claim to agent tasks.

Google deprecated the deep-research-pro-preview-12-2025 agent. Google said the agent would be shut down on 2026-10-23.

Google's API changelog marked gemini-3.7-flash deprecated and replaced by gemini-3.8-flash, and gemini-3.5-flash deprecated and replaced by gemini-3.6-flash. Requests to the retired model strings are routed to the replacements automatically.

Google deprecated the gemini-3.1-flash-image model. Google announced no shutdown date for the model.

Google released Gemini Nano Banana 2.1, identified as a high-efficiency image-generation and conversational-editing model. Google described it as the more efficient counterpart to Gemini 3 Pro Image and listed standard image-output pricing at $30 per million tokens.

Google set Gemini 4 Argon's introductory price at $2 per million input tokens and $10 per million output tokens. Cached input tokens are priced at 95% off the input-token price during the introductory period.

Google announced Gemini 4 Argon and began rolling it out to trusted cyber defenders. Initial access is through Google's Fairwind Program.

Google made Gemini 3.8 Live with Live Avatar available in Gemini Enterprise. Live Avatar combines Gemini 3.8 Live dialogue with streaming video to create a visual persona.

Google made Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS generally available. It also released the Gemini API Voices endpoint.

Google limited Gemini 2.5 model access to users who had actively used the models previously. Google directed new projects to Gemini 3.5 Flash-Lite or Gemini 3.8 Flash.

Reuters and Bloomberg reported that banks were arranging a $22 billion loan for Crux AI. Reuters reported that the debt would be used to purchase Google TPUs, while Bloomberg reported it could later be replaced with longer-term institutional financing.

Google published pricing of $0.005 per minute for audio input and $0.018 per minute for audio output for Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. The models are available through the Live API.

Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking to general availability. The production Live API endpoints are gemini-3.8-live and gemini-3.8-live-extended-thinking.

Google released Lyria 3.5 as a generally available music-generation model in the Gemini API. The model generates full-length songs and provides controls for duration and structure.

Broadcom said it began production shipments of the next-generation TPU 8i for Google. The statement was made during Broadcom's fiscal third-quarter earnings call.

Google introduced Gemini 3.8 Flash Cyber as a cybersecurity model for vulnerability detection and automated patching. The model was made available to trusted defenders through the Fairwind Program.

Broadcom's AI semiconductor revenue more than tripled to $16.7 billion with custom accelerators 73 percent of it, and Hock Tan laid out the customer roadmap: multi-tens of billions of dollars of TPUs a year for Google, 1 GW of Ironwood for Anthropic in 2026 then 5 GW of TPU 8i in 2027 and 10 GW more in 2028, 1.3 GW of Jalapeño for OpenAI in 2027 with a second generation near tape-out and a third in development, and MTIA production shipments for Meta. Guidance: $21.7 billion of AI revenue in Q4, $58 billion for fiscal 2026 and about $115 billion in fiscal 2027.

Google released agentic video understanding for Gemini 3.7 Flash, Gemini 3.6 Flash, and Gemini 3.5 Flash-Lite. The feature was available across the Interactions and GenerateContent APIs.

Data Center Dynamics reported that the Google and Blackstone neocloud offering TPU services launched as Crux AI. Bloomberg reported that Blackstone provided $5 billion and that Crux AI aims to rent Google AI processors to customers in 2027.

Google released gemini-omni-1.1-flash as the generally available version of Gemini Omni Flash. Google described it as a conversational video-generation and editing model.

Google released Gemini Robotics ER 2 preview endpoints for robotics. Google DeepMind said Gemini Robotics ER 2 was available on Google AI Studio and in private preview on Gemini Enterprise Agent Platform.

Google released gemini-3.1-flash-lite-image, called Nano Banana 2 Lite, to general availability. Google described it as optimized for low-latency and cost-effective image generation and editing.

Google released gemini-omni-flash-preview in public preview. Google described it as a model for high-speed video generation and conversational video editing.

Google launched public-preview support for the Computer Use tool in Gemini 3.5 Flash. The support was recorded in the Gemini API changelog.

Google added streaming support for Gemini 3.1 Flash TTS Preview. Streaming was supported through streamGenerateContent and the Interactions API.

Google released Gemini 3.5 Live Translate. Google described it as an audio model for live speech-to-speech translation.

Google shut down Gemini 3.1 Flash-Lite Preview. The shutdown applied to the gemini-3.1-flash-lite-preview endpoint.

Google released gemini-3.1-flash-lite as the generally available Gemini 3.1 Flash-Lite version. Google described it as optimized for speed, scale, and cost efficiency.

Google shut down the Gemini Robotics ER 1.5 Preview model. The shutdown applied to the gemini-robotics-er-1.5-preview endpoint.

Google previewed its first two-chip TPU generation: 8t for large-scale training and 8i for low-latency agentic inference, with general availability promised for later in 2026 for the first parts. Reports identified Broadcom as the design partner on 8t and MediaTek on 8i, both on TSMC 2nm, and Broadcom later said Anthropic would take 5 GW of 8i in 2027.

Google launched Gemini 3.1 Flash TTS Preview. Google described it as a text-to-speech model.

Google released Gemini Robotics ER 1.6 Preview. Google described it as an updated robotics model.

Anthropic called it its most significant compute commitment to date: multiple gigawatts of next-generation TPU capacity from 2027, mostly sited in the United States, on top of the million-TPU deal of October 2025. Broadcom's 8-K put the figure at 3.5 GW and disclosed a long-term supply agreement with Google for future TPU generations through 2031. Anthropic's run-rate revenue had passed $30 billion.

Google introduced Flex and Priority inference tiers for the Gemini API. Google said the tiers provided options for optimizing cost or latency.

Google launched Veo 3.1 Lite Preview under the endpoint identifier veo-3.1-lite-generate-preview. Google said it was designed for rapid iteration and high-volume applications.

Google released Gemini 3.1 Flash Live Preview. Google described it as an audio-to-audio model for real-time dialogue and voice-first AI applications.

Google launched Lyria 3 Clip and Lyria 3 Pro music-generation models. Lyria 3 Clip generated 30-second clips, while Lyria 3 Pro generated full-length songs.

Google launched Gemini 3.1 Flash-Lite Preview. Google described it as the first Flash-Lite model in the Gemini 3 series.

Three months after the first report, The Information said Meta had agreed to a multibillion-dollar deal to rent TPUs from Google Cloud to train and run its next-generation models, alongside multibillion-dollar purchases of Nvidia's Vera Rubin and AMD's MI400 GPUs announced the same month. Meta was buying from every accelerator vendor at once while its own MTIA program continued.

Google released gemini-2.5-flash-native-audio-preview-12-2025. The model was a native-audio model for the Live API.

Google launched enhancements to Gemini 2.5 Flash TTS preview and Gemini 2.5 Pro TTS preview. Google listed expressivity, pacing, and dialogue among the changes.

The Information reported that Meta was negotiating to deploy TPUs in its own data centers in 2027 and to rent them from Google Cloud in 2026, and that Google was now promoting TPUs for installation in customers' facilities rather than only in its cloud. Alphabet rose 4 percent and Nvidia fell as much as 7 percent on the report; a Google spokesperson said demand for both TPUs and Nvidia GPUs was accelerating.

Google Cloud said Ironwood would be generally available within weeks, with 10x the peak performance of TPU v5p and more than 4x Trillium per chip, and introduced Arm-based Axion N4A instances. The post opened by listing Anthropic's Claude with Gemini, Veo and Imagen as models that train and serve on TPUs.

Google reduced the image input-token count for Gemini 2.5 Flash Image from 1,290 to 258. Google said the reduction lowered the cost of image editing.

Anthropic announced a deal worth tens of billions of dollars to expand its use of Google Cloud TPUs to as many as one million chips, bringing well over a gigawatt online in 2026, while keeping Amazon as its primary training partner. It was the first time a frontier lab had committed to TPUs at Nvidia-order scale, and it made Anthropic the anchor customer for Ironwood.

Google released Veo 3.1 and Veo 3.1 Fast in public preview. Google said the models included new features.

Google launched Gemini 2.5 Computer Use Preview. The launch was recorded in the Gemini API changelog.

Google launched Gemini 2.5 Flash Image GA. The launch made the image-generation model generally available.

Google shut down Gemini 1.5 API models. The release notes identified the Gemini 1.5 models as shut down.

Google released Gemini Robotics ER 1.5 in preview. The release identified it as a model.

Google released gemini-2.5-flash-native-audio-preview-09-2025 for the Live API. Google said the model included improved function calling and speech cutoff handling.

Google launched Veo 3 and Veo 3 Fast in general availability. The release included lower pricing and options for aspect ratios, resolution, and seeding.

Google launched Gemini 2.5 Image Preview. Google described it as a native image-generation model.

Google made Deep Think available in the Gemini app to Google AI Ultra subscribers. The availability was announced by Google Blog.

Google released gemini-2.5-flash-lite. Google described it as a Gemini 2.5 model for speed, low cost, and performance.

Google released Gemini 2.5 Pro and Gemini 2.5 Flash as stable and generally available models. The release was announced through Google Blog.

Google released Gemini 2.5 Flash TTS and Gemini 2.5 Pro TTS preview models. The models could generate speech with one or two speakers.

Google released gemini-2.0-flash-preview-image-generation as a preview model. The model generated and edited images.

Google released gemini-2.5-flash-preview-04-17. Google described the preview model as optimized for price-performance and adaptive thinking.

Google released gemini-2.0-flash-live-001 as a public-preview Live API model. Billing was enabled for the release.

At Cloud Next Google introduced Ironwood as the first TPU designed specifically for inference and 'thinking' models, with 192 GB of HBM per chip and pods of 9,216 chips delivering 42.5 exaflops. It became the chip behind Google's TPU sales push to Anthropic and Meta over the following year.

Reuters, citing The Information, reported that Google was preparing to work with MediaTek on a next-generation TPU expected to be manufactured in 2026, alongside its long-standing Broadcom partnership. Google, MediaTek and Broadcom did not comment. The two-partner split later surfaced as TPU 8t from Broadcom and TPU 8i from MediaTek.

Google launched an experimental Gemini 2.0 Flash model capable of image generation and editing. The model was recorded in the Gemini API changelog.

Google released gemini-2.0-flash-lite as a generally available Gemini 2.0 Flash-Lite model. Google described the model as optimized for speed, scale, and cost efficiency.

Google ended support for Gemini 1.0 Pro. The status was recorded in the Gemini API release notes.

Google released an experimental version of Gemini 2.0 Pro. Google described it as a model for coding performance and complex prompts.

Google made the updated Gemini 2.0 Flash generally available through the Gemini API in Google AI Studio and Vertex AI. The release was announced by Google Blog.

Google released Gemini 2.0 Flash Thinking Mode in public preview. The release was listed in the Gemini API changelog.

Google released gemini-1.5-flash-8b-001 as a stable model. Google described it as its smallest Gemini API model.

Google released stable Gemini 1.5 Pro 002 and Gemini 1.5 Flash 002 versions for general availability. The releases were identified as gemini-1.5-pro-002 and gemini-1.5-flash-002.

Google removed support for Gemini 1.0 Pro Vision from Google AI services and tools. The removal affected the Gemini API model line.

Google made the two-million-token context window for Gemini 1.5 Pro generally available. The capability had previously been offered through a waitlist.

Google made gemini-1.5-flash-001 generally available. This was a production Gemini API release for Gemini 1.5 Flash.

Google made gemini-1.5-pro-001 generally available. This was a production Gemini API release for Gemini 1.5 Pro.

Google introduced a two-million-token context window for Gemini 1.5 Pro through a waitlist. The capability was listed in the Gemini API release notes.

Google announced Trillium with a 4.7x jump in peak compute per chip over TPU v5e, doubled HBM and interconnect bandwidth and 67 percent better energy efficiency, and said the models unveiled at I/O, including Gemini 1.5 Flash, were trained and served on TPUs. Trillium was the last single-purpose TPU before the split into inference and training parts.

Google's Gemini API release notes record Gemini 1.5 Flash as released in preview. A Google Blog post dated May 14 discussed applications requiring lower latency and lower serving cost.

Google released Gemini 1.5 Pro in preview through the gemini-1.5-pro-latest endpoint. The release made the model available through the Gemini API.

Sundar Pichai and Demis Hassabis presented Gemini as Google's most capable and general model, built natively multimodal and shipped in three sizes so the same family could run from a data centre to a phone.

The same day it announced Gemini, Google Cloud introduced TPU v5p, its most powerful accelerator to date, with 8,960 chips per pod, more than twice the FLOPS and three times the HBM of TPU v4, and said Gemini had been trained on and was served from TPUs. It set the pattern for the next three years: each Google model generation ships on a new TPU.