Google Opens Gemini 4 Argon to Paying Users at $2 Per Million Input Tokens

Google opened Gemini 4 Argon to paying users worldwide on Tuesday, bringing its flagship model to the Gemini app, Search and developer tools. The introductory API rate starts at $2 per million input tokens. The introductory rate doubles after November 30, 2026.

Access depends on the product

The October 13 rollout gives Google AI Pro and Ultra subscribers access in the Gemini app and AI Mode in Search. Argon in the Antigravity desktop app and command-line tool requires Ultra. Pro subscribers can generate code through the Gemini app's Canvas mode.

Google's announcement offers developer access through the paid Gemini API tier, while access inside AI Studio and Android Studio as subscription products is reported as requiring AI Ultra. Enterprises can use Argon through Gemini for business, the API and Antigravity. Free and Plus subscribers are excluded from the announced consumer rollout. Google has announced no date for Argon access for free or Plus users.

The release follows the September 30 unveiling, when access was restricted to selected cybersecurity partners in Google's Fairwind program. That program is also expanding to hundreds of additional partners.

The launch rate expires in November

At launch, API output costs $10 per million tokens. Cached input, material reused from an earlier request, receives a 95% discount, putting the introductory cached-input rate at $0.10 per million tokens.

The promotional window ends on November 30, 2026. After that, ordinary input costs $4 per million tokens and output costs $20 per million, doubling both launch rates.

The API has a unified input-output limit of one million tokens. That allowance covers material sent to the model and generated responses together. A continuation feature pauses long responses and resumes them across follow-up calls, allowing extended reasoning without a single request running into a timeout.

Cheaper tokens can buy a more expensive task

Google's launch announcement puts Argon first on the Vals Index, which measures professional work across finance, coding, legal and tax tasks. On the Artificial Analysis Intelligence Index, Argon is level with two rivals and behind two others. Argon scored 53 on the Artificial Analysis Intelligence Index, level with GPT-6 Astra and Claude Fable 5.1, behind Claude Opus 5.5 at 58 and Claude Sonnet 5.5 at 56.

In the Artificial Analysis results available at the October 13 rollout, Argon averaged about 62,000 output tokens per Intelligence Index task, against 27,000 for GPT-6 Astra at its maximum reasoning setting, about 2.3 times as many. Argon was tested at high reasoning, its highest available setting. Both models were tested at their highest available reasoning settings, which may not be equivalent.

At introductory rates, the measured cost per task was $1.99 for Argon, against $3.26 for Astra. Applying Argon's standard rates to that workload raises its cost to $3.98, above Astra.

The same independent evaluation put Argon at 57% on Terminal Bench 4, behind Claude Sonnet 5.5 at 64%, Claude Opus 5.5 at 60% and Astra at 59%.

Internal engineering results remain unverified

Google's October launch examples include a Chrome refactor separating webpage handling from legacy browser-feature code across hundreds of files. The work is intended to let live pages move between tabs, side panels and floating windows without reloading.

At YouTube, Argon identified repeated emoji rendering and helped engineers cut the comments frontend's CPU footprint in half. The launch account describes reductions in server comment-load latency of 30 milliseconds and client end-to-end latency of 20 milliseconds.

For its open Gemma model, Google reported a 2.85-fold inference speedup during work over the preceding few days. On an Nvidia A100 running vLLM, throughput rose from about 37,900 to more than 108,000 tokens per second for Gemma 4 E2B, with identical token output.

The productivity and speed claims rest on Google's own measurement and have not been independently verified in aggregate, though some underlying code changes are public.

Meta Releases 30B Open-Weight Muse Glimmer and Promises Spark 1.2 Weights

Meta released Muse Glimmer, a 30-billion-parameter open-weight model, Monday. Glimmer distills Muse Spark to run local agents on a single high-end Mac or PC. Meta promised Muse Spark 1.2 weights in co

The Implicator

Meta Opens Muse Glimmer as Zuckerberg Presses U.S. AI Policy Shift

On Aug. 10, 2026, Meta released Muse Glimmer, a 30-billion-parameter model whose weights developers can download. It was built for agents that run on a Mac or PC. The release carried a policy argumen

The Implicator

Alibaba and DeepSeek Release 2.4T and 1.6T Models for Cloud-Scale AI

On Aug. 12, Alibaba finally published the Qwen3.8 checkpoint it had promised in July, a model carrying 2.4 trillion parameters. Nvidia put the files on its newest full-rack Blackwell system and publis

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai