Google opens Gemini 4 Argon to all users at half its list price, but independent rankings put it behind Claude Opus 5.5

Argon has a 1-million-token context window and a 1-million-token output limit, and it leads Google's own tables on knowledge work and long context while finishing last of four models on two agentic coding benchmarks.

On October 13, Google opened Gemini 4 Argon to developers, enterprises, scientists and consumers. Google announced the model on September 30 and at first limited it to vetted cybersecurity partners in its Fairwind program. The Russian-language channel AI ML Big Data summarized Google's stated targets as long multistep tasks in software development, financial analysis, scientific research and cybersecurity. The DeepMind launch post gives agentic workflows equal emphasis.

The model accepts text and image input, has a 1-million-token context window, and has a 1-million-token output limit. Earlier Gemini models capped output at 64,000 tokens. Google set introductory pricing at $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 when the promotion ends. Cached input costs $0.10 per million tokens during the promotion and $0.20 afterward, according to a pricing breakdown by eesel.

Google's own comparison table, tabulated by MindStudio, compares Argon with OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 and Claude Opus 5.5. Argon wins 13 of 19 rows, ties one and loses five. Its widest leads are 68.9% on the Vals Index and 51.3% on AutomationBench, and it scores 84.2% on GraphWalks between 256,000 and 1 million tokens. It finishes last of the four on FrontierSWE v2 and Terminal-Bench 4.0. Google chose the rows in that table, so a 13-of-19 count measures Google's selection as much as the model.

An independent benchmark tells a different story. On the Artificial Analysis Intelligence Index, Argon at high reasoning scores 53, roughly level with GPT 6.1 Astra and behind Claude Opus 5.5 (58) and Claude Sonnet 5.5 (56). Third-party write-ups also report a hallucination rate of about 15%, below many current frontier models, but this desk has not seen an independent reproduction of that figure. The supported capability gap is narrower than the launch framing suggests: Argon is a long-context, document-heavy knowledge-work model, and it does not lead on agentic coding.

What this means

Google is competing on price, context length and distribution through Workspace, Cloud and Android, because it cannot yet claim the top general or coding score. The exposed parties are OpenAI's GPT-6 Astra and Anthropic's enterprise document and analysis business, where long-context accuracy and a 1-million-token output limit translate directly into fewer chained calls. Anthropic's coding-agent revenue is less exposed, because Argon trails on Terminal-Bench 4.0 and FrontierSWE v2. The market already expects a strong Google model. The new element is that Google launched below its own list price while it trails on an independent index. Meta did the same in July when it priced Muse Spark 1.1 at about a quarter of rival rates, so this is now a pattern among platform companies. Consensus coverage counts benchmark wins, but the commercial contest is now over cost per completed task. Lower prices compress per-token revenue for every closed-model vendor, and Alphabet accepts a lower cloud AI margin in exchange for share. The likelier outcome is that OpenAI and Anthropic answer within weeks with price cuts or long-context discounts on their mid and top tiers rather than with new flagship models. That depends on independent long-context tests confirming Argon's GraphWalks and Vals results. The read is wrong if Google extends the promotional price indefinitely without gaining measurable share, which would show that price alone does not move enterprise buyers.

Part of a tracked trend

Hyperscalers Enter the Frontier API Market

Large platform companies increasingly convert in-house frontier models into metered, incumbent-compatible APIs, competing on switching cost and distribution rather than pure capability.

Townsquare

Start a discussion in Townsquare.

Comments

No comments yet.