Two weeks after the announcement, you can actually use Google's new top model. It sits straight at the top of the LMArena text leaderboard, but the low price only lasts for a short time. Meanwhile, Workday and Anthropic are linking their AI agents together. Here is what you need to know.
Gemini 4 Argon: available since 13 October
Google announced Argon on 30 September (our piece on it). Since 13 October, the model has been according to Google available worldwide:
- via the Gemini API and AI Studio, and for organisations via Vertex AI;
- in the Gemini app and AI Mode in Search, if you have Google AI Pro or Ultra;
- in Google Antigravity, for Ultra subscribers.
The context window is 1 million tokens, with input and output together falling within that limit. Google aims the model at legal work, finance, science and programming.
The price will soon double. During the introductory period you pay $2 per million input tokens and $10 per million output tokens. From 30 November, that becomes $4 and $20. In our OpenRouter data, Claude Sonnet 5.5 now also costs $2/$10, and Claude Opus 5.5 $4/$20. After November, you therefore pay for Argon the same as for Opus 5.5. For a business case, calculate with the second rate.
How good is it? Google says Argon ranks number 1 on the Vals Index, a test with tasks from finance, legal work, tax work and programming. That is a claim by Google itself. More independent is LMArena, where users weigh anonymous answers against each other. In our arena data of 13 October, Argon ("high" setting) tops the text leaderboard with 1525 points. Next come Claude Opus 4.6 (1505) and Claude Fable 5 and Opus 5.5 (both 1504). Argon has only 4,932 votes and a margin of error of 9 points, so the lead could still shrink. On the code leaderboard, Argon does not appear in the top 5; there Claude Opus 5.5 leads with 1815 points. You can find all scores at /modellen/.
Privacy. The announcement says nothing about your data. The Gemini API terms, however, are clear. If you live or work in the EEA, the rules for paid services always apply, including for AI Studio and free API use. Google then does not use your prompts and answers to improve products. They are only kept for a limited time to detect misuse. Different terms apply to the Gemini app. For how Gemini compares with other models, see /gemini/ and /veiligheid/.
Workday and Claude: HR and finance work from a single chat
On 13 October, Workday and Anthropic announced a partnership. Claude and Workday's own agent Sana exchange data and tasks via the Model Context Protocol (MCP). With the "Sana in Claude" plugin, you can analyse HR and finance data from Workday within Claude and carry out actions on it. According to Workday, every action follows your organisation's existing permissions and audit settings.
The plugin is currently in early access. General availability is planned for the end of 2026. The press release says nothing about storage in the EU. If you work in HR, financial advice or at a bank, ask before use which Workday data reaches Claude through the connection and which data processing agreements apply. For how Claude handles business data, read the model page.
In brief
- Gemini 4 Argon is available via the API, Vertex AI and the Gemini app (Pro/Ultra). Until 30 November it costs $2/$10 per million tokens, after that $4/$20.
- On LMArena, Argon now tops the text leaderboard (1525), but with few votes. For code, Claude Opus 5.5 leads.
- Via the API, Google does not train on your data in the EEA.
- Workday data will soon be usable from Claude. Still in early access, generally available at the end of 2026.