Gemini 4 Argon is Google's flagship reasoning model, generally available on the Gemini API since October 13, 2026 under the model ID gemini-4-argon. Google describes it as its most capable model for complex coding, agentic, math, and science tasks. Its introductory API price is $2 per million input tokens and $10 per million output tokens, and that price doubles on December 1, 2026.
This guide covers what Argon costs, how its thinking levels work, which Slack work it is worth the price for, and how to run it in a Slack agent with Runbear.
Gemini 4 Argon at a glance
| Spec | Gemini 4 Argon |
| Model ID | gemini-4-argon (Stable) |
| Released | October 13, 2026 (GA on the Gemini API) |
| Context window | 1,048,576 tokens, shared by input, thinking, and output |
| Default max output | 65,536 tokens (longer output needs background mode) |
| Inputs / output | Text, image, video, audio, PDF in; text out |
| Thinking | Levels low, medium (default), high; no off level listed |
| Price through Nov 30, 2026 | $2 input / $10 output per 1M tokens |
| Price from Dec 1, 2026 | $4 input / $20 output per 1M tokens |
| Caching | Implicit only, $0.10 per 1M cached tokens ($0.20 from Dec 1) |
| Free tier | Not available |
Figures as of October 2026, from Google's Gemini API model page, pricing page, and changelog.
What it costs, and the December 1 price change
On the Gemini API Standard tier, Argon lists at $2 per million input tokens and $10 per million output tokens through November 30, 2026. From December 1, 2026 the same tokens cost $4 and $20. Output pricing includes thinking tokens, so a request that thinks longer costs more even when the visible answer is short.
There is no long-context surcharge: unlike Gemini 3.1 Pro, Argon's price does not change above 200,000 prompt tokens. Batch and Flex tiers are half the Standard price, Priority costs more, and grounding with Google Search is billed per search query after a monthly free allowance. Argon is not on the free tier.
To see what that means in practice, take 1,000 requests that each send 20,000 input tokens and produce 2,000 output tokens. At list price with no caching, that is $60 on Argon today, $120 on Argon after December 1, $64 on Gemini 3.1 Pro Preview, and $22.50 on Gemini 3.8 Flash. Real token counts depend on the task and the thinking level, and Google notes that Argon can spend more of its output budget on thinking, so measure on your own requests.
How thinking works on Argon
Google's documentation lists three thinking levels for Argon, low, medium, and high, with medium as the default, and no level that turns thinking off. Low uses the fewest tokens and answers fastest, but Google warns quality can drop on hard tasks. High uses the most tokens and has the slowest first token. Google's advice is to test all three and pick the lowest level that meets your quality bar.
Two more migration notes from Google matter for chat use. Temperature, top_p, and top_k are ignored on Argon. And explicit context caching is not supported; Argon only uses implicit caching, which applies automatically to prompts of at least 4,096 tokens.
What Gemini 4 Argon is good for in Slack
Argon is priced for hard problems, not for every message in a busy channel. In Slack, that points to agents that get fewer, harder questions. This is our guidance, not Google's.
1. Engineering design and code questions
Google positions Argon for complex coding and agentic tasks. A Slack agent that architects and engineers @mention with design trade-offs, failure modes, or a tricky bug is the kind of low-volume, high-value use where a stronger model pays for itself.
2. Analysis that needs several steps of reasoning
Questions like "which of these three vendor contracts has the riskiest renewal terms" or "walk me through why last month's numbers don't add up" need the model to hold several facts at once and reason across them. The 1M-token context window also leaves room for long documents in a single request.
3. Math and science questions
Google names math and science among Argon's strengths. Research, data, and finance teams that ask quantitative questions in Slack are a better fit than a general help channel.
When to pick a cheaper model instead
For routine answers from your docs, triage, and short summaries, a Flash or Flash-Lite model is usually enough and costs a fraction of Argon. One setup is two agents: a fast, low-cost one for the everyday channel and an Argon agent for the channel where hard questions go.
Gemini 4 Argon vs Gemini 3.1 Pro vs Gemini 3.8 Flash
| Gemini 4 Argon | Gemini 3.1 Pro Preview | Gemini 3.8 Flash | |
| Input / output per 1M (Standard) | $2 / $10 through Nov 30; $4 / $20 from Dec 1 | $2 / $12 up to 200K prompt; $4 / $18 above | $0.75 / $3.75 through Dec 31, 2026; $1.50 / $7.50 from Jan 1, 2027 |
| How Google describes it | Most capable model for complex coding, agentic, math, and science tasks | 3rd-generation Pro model for multimodal understanding, agentic capabilities, and vibe-coding | Most intelligent Flash model, for long-horizon software engineering and autonomous agents |
| In Runbear | Gemini agents (Create manually) and the API | Claude Agent SDK agents, Gemini agents, and the API | Claude Agent SDK agents |
| Our pick for | Low-volume, hard questions | Long-context reasoning on the default runtime | Everyday team questions |
Prices as of October 2026 from the Gemini API pricing page. The last row is our guidance, not Google's.
How to run Gemini 4 Argon in a Slack agent with Runbear
As of October 2026, Runbear offers Gemini 4 Argon on Gemini agents, not on the Claude Agent SDK runtime that new agents usually get. Claude Agent SDK agents offer Gemini 3.1 Pro, Gemini 3.8 Flash, and Gemini 3.5 Flash Lite alongside Claude and GPT models.
- Go to Agents, select Create Agent, then choose Create manually at the bottom of the dialog.
- Pick Google Gemini.
- In the Model field, choose gemini-4-argon.
- Add your instructions and knowledge sources, then add the agent to a Slack channel. Our step-by-step guide on how to add an AI agent to a Slack channel walks through the Slack side.
You can also create a Gemini agent through the Runbear API by sending provider type gemini with the model gemini-4-argon. On Runbear-provided access, usage is billed in Runbear credits, and the credit rate follows the model you pick.
FAQ
Can I use Gemini 4 Argon in Slack?
Yes, through a Slack agent or app that lets you choose the model. In Runbear, create a Gemini agent with Create manually, choose gemini-4-argon in the Model field, and add the agent to a Slack channel.
How much does Gemini 4 Argon cost?
On the Gemini API Standard tier it costs $2 per million input tokens and $10 per million output tokens through November 30, 2026, and $4 and $20 from December 1, 2026. Output pricing includes thinking tokens. It is not available on the free tier.
Can I turn off thinking on Gemini 4 Argon?
Google's documentation lists only low, medium, and high thinking levels for Argon, with medium as the default, and no option to turn thinking off. Use low when you need the fastest, cheapest answers.
Is Gemini 4 Argon better than Gemini 3.1 Pro?
Google calls Argon its most capable model for complex coding, agentic, math, and science tasks. Until November 30 its input price matches 3.1 Pro and its output price is lower. From December 1 Argon's price doubles to $4 and $20, so its output then costs more than 3.1 Pro's. Compare cost per task on your own questions, since Argon can use more thinking tokens.
Runbear runs shared AI agents in Slack and Microsoft Teams that your team can build without code. See how teams use shared agents, compare the best AI agents for Slack, or read our guide to Claude Haiku 5.5 in Slack.