Google DeepMind

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 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.

Google positioned it for long-horizon software engineering, autonomous agents and complex enterprise workflows — the fourth numbered Flash model since May, still with no Pro release in the 3.5 line.

Two dedicated transcription models, one batch and one streaming, were built on Gemini's audio understanding rather than sold as a general model feature.

Google called it its most intelligent workhorse model yet for coding and agents, and put it on an introductory price good to the end of the year.

TechCrunch noted what the July release did not contain, and pointed at a Bloomberg report the previous week that Google was facing internal delays on 3.5 Pro after it fell short of its own targets. Pro is Google's high-capability tier; the gap left the generation's frontier slot filled by Flash models.

Google shipped stable versions of two more Flash models, 3.6 Flash cheaper than the 3.5 Flash it followed. The Pro model the 3.5 family had been announced around was absent.

The generation that introduced the agentic framing was switched off, with developers pointed at 3.5 Flash or 3.1 Flash-Lite.

Gemini 3.1 Flash Image and Gemini 3 Pro Image left preview together, keeping the Nano Banana names that users had given them.

Google called it its most intelligent model for sustained frontier performance on agentic and coding tasks — a claim being made for a Flash model, with no 3.5 Pro alongside it.

Under the model lifecycle policy Google had launched in January, the preview endpoint that opened the generation was retired less than four months after it appeared.

The image line got its high-efficiency variant, aimed at speed and high-volume use rather than maximum quality.

Three months after Gemini 3 Pro, Google shipped the first iteration of the Pro tier.

Developers who pinned gemini-pro-latest or gemini-flash-latest were moved onto the third generation without changing a line of code.

Google said the Flash model delivered frontier-class performance rivalling larger models at a fraction of the cost, a month after the Pro launch.

Google introduced a unified API surface for Gemini models and agents, the plumbing that the agentic releases through 2026 would be built on.

Two days after the Pro model, the image model that the internet had renamed Nano Banana got its Gemini 3 iteration.

The generation opened with the Pro tier in preview, pitched on reasoning, multimodal understanding, agentic work and coding.

Google described 2.5 as thinking models that reason through a problem before answering, and led with 2.5 Pro Experimental. The framing matched the reasoning turn the rest of the industry had taken.

Google put its workhorse tier first for a new generation, pairing 2.0 Flash with agent prototypes and native tool use. From here the Flash models, not Pro, set the release cadence.

The first 1.5-series model went to early testers with a mixture-of-experts architecture and, Google said, the longest context window of any large-scale foundation model at the time. Long context became the line's defining claim.

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.