← Model indexMODEL PROFILE / gemini-4-argon

Google / current

Gemini 4 Argon

Google's flagship reasoning model with a unified one-million-token context and one-million-token decoding.

Context1M
ModalityText · Image · Video · Audio
AccessAPI · Cloud
ReasoningControllable

Where it fits.

Built for complex software engineering, deep mathematical and scientific problem solving, and autonomous multi-step agents. Announced September 30, 2026 and broadly available October 13. It offers a unified 1M-token input window shared with up to 1M-token decoding, adjustable thinking levels low/medium/high with medium as the default, and works across text, image, video and audio.

02

Core capabilities

  • Unified 1M input context with up to 1M decoding
  • Adjustable thinking levels from low through high
  • Text, image, video and audio understanding
  • DeepSWE score of 77.9% on Google's reported benchmark
03

Best-fit workflows

  • Complex refactoring
  • Scientific and mathematical work
  • Long-horizon agents
04 / OPERATIONS

From evaluation
to production.

Deployment notes

Budget at the post-December rate. Pin the stable gemini-4-argon ID, cap output tokens unless long generation is required, and log thinking-token spend per task to keep agent workflows predictable.

Watch before choosing

Introductory pricing doubles on December 1, 2026; budget pilots at the higher rate from the start. Not available on the free tier. Only implicit caching is supported, and thinking tokens are billed as output, so long reasoning runs can dominate cost. Pricing rises to $4 input and $20 output per million from December 1, 2026.

05

Family & versions

Specifications and positioning checked against provider documentation.

Open official source↗