Gemini 4 Argon
Language Available Full comparison ↗ google-deepmind/gemini-4-argon · by Google DeepMind
· decoder-only-transformer
Pricing — 1 offering(s)
About this offering: Gemini 4 Argon, Google's flagship reasoning model, generally available on the Gemini API since 2026-10-13 (paid tier only; not on the Free tier or the Google AI Pro plan). One price at every prompt length, with no >200K long-context band. Introductory pricing runs through 2026-11-30.
Input tokens
- $2.00 / 1M tokens (input)
Current
2026-10-14 → 2026-11-30
Introductory rate, $2.00/1M input through 2026-11-30; $4.00/1M from 2026-12-01 per Google's pricing page.
- $4.00 / 1M tokens (input) Future 2026-12-01
Cached input tokens
- $0.10 / 1M tokens (cached input)
Current
2026-10-14 → 2026-11-30
Implicit caching only (explicit caching is not supported); 95% off input. Introductory rate, $0.10/1M through…
Implicit caching only (explicit caching is not supported); 95% off input. Introductory rate, $0.10/1M through 2026-11-30; $0.20/1M from 2026-12-01.
- $0.20 / 1M tokens (cached input) Future 2026-12-01
Output tokens
- $10.00 / 1M tokens (output)
Current
2026-10-14 → 2026-11-30
Includes thinking tokens. Introductory rate, $10.00/1M output through 2026-11-30; $20.00/1M from 2026-12-01…
Includes thinking tokens. Introductory rate, $10.00/1M output through 2026-11-30; $20.00/1M from 2026-12-01 per Google's pricing page.
- $20.00 / 1M tokens (output) Future 2026-12-01
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Gemini 4 Argon fits into a cost-aware routing setup
See how →Capability profile
How Gemini 4 Argon rates across core capability dimensions, with the task-level evidence behind each rating.
Google's model page positions it for "deep mathematical and scientific problem solving", with tunable thinking levels (low, medium, high).
Google's model page: "built for complex software engineering"; code execution supported. No independent coding benchmark recorded yet.
Function calling, code execution, file search, Google Search/Maps grounding, URL context and computer use (Preview) are listed as supported (ai.google.dev/gemini-api/docs/models/gemini-4-argon).
1,048,576-token unified input and output window; 65,536 output tokens by default, up to about 1M with long decode.
Google's migration guide says hard prompts can take several minutes and recommends client timeouts of at least 15 minutes.
$2/$10 per 1M introductory pricing through 2026-11-30, then $4/$20; implicit caching cuts cached input by 95%.
Ratings are estimated — limited independent data is available for this model.
Operator guidance
Route here for the hardest reasoning and agentic-coding tasks where latency isn't a constraint. Introductory pricing ($2/$10) undercuts most frontier peers until 2026-11-30; after that it's $4/$20, so re-check routing decisions in December. Explicit context caching isn't supported, so prompts relying on cachedContents need migrating.
Use cases
- The hardest coding, maths and science problems, where depth matters more than latency
- Autonomous multi-step agents, including Google's Antigravity managed agent
- Very long generations (past 64K tokens, up to about 1M) via long decode
Limitations
- Not available on the Free tier or the Google AI Pro plan
- Implicit caching only; explicit context caching is not supported
- Sampling parameters (temperature, top_p, top_k) are ignored
- Slow on hard prompts (minutes per response)
- Capability ratings are qualitative, from Google's own materials, not a cited benchmark run