Gemini 4 Argon handles deep multi-step reasoning with configurable thinking levels and can generate up to about one million output tokens. Google lists its retirement date as October 2027 or later.
Gemini 4 Argon
Google model for multi-step reasoning with configurable thinking
Checked 2026-11-12 against Google's documentation
| Provider | |
| Released | 2026-10-13 |
| Status | Current |
| Context window | — |
| Pricing | Paid |
| Capabilities | textvisioncodereasoning |
| API docs | Docs ↗ |
Benchmarks
Benchmark data not yet available for this model.
Compare with similar models
| Gemini 4 Argon | Claude Opus 4.6 | Gemini 2.5 Pro | Claude Fable 5.2 | |
|---|---|---|---|---|
| Context | — | 200K | 1.05M | 1M |
| MMLU | — | 92.0 | 91.0 | — |
| HumanEval | — | 94.5 | — | — |
| MATH | — | 97.1 | 97.0 | — |
| GPQA | — | 79.9 | 84.0 | — |
| Pricing | Paid | Paid | Paid | Paid |
| Input $/M | — | $15.00 | $1.25 | $8.00 |