Releases

Gemini 4 Argon

MiniRouter and Gemini 4 Argon by Google, separated by a silver fold of light on black.

Google's new reasoning model is built for sustained coding and complex professional work. Explore the evidence, introductory pricing and current API availability.

Gemini 4 Argon is available on MiniRouter as . The model page has current pricing, provider routes and API setup.

View Gemini 4 Argon

More room for hard problems

Context
1M tokens
Vercel's published context window.
Best fit
Sustained reasoning
Code migrations, research and multi-step business tasks.
API examples
Text requests
Vercel advertises multimodal input; the setup here starts with text.
Thinking
Low · medium · high
Vercel's advertised reasoning settings. Output limits include thinking.

MiniRouter currently supports text input for this model. Start with the text API setup →

Built for the longer task

Bring the whole problem

Pair your instructions with code and source material.

Your application supplies tools and checks their results. The advertised token ceiling is not a guarantee that every request can run to that length.

Coding and knowledge work

Three Google-reported results. Each measures a different task; these are not MiniRouter performance tests.

DeepSWE v1.1

  • Gemini 4 Argon77.9%

0–100% · higher is better

Long-horizon software engineering in the mini-swe-agent harness. Google-computed, pass@1 at highest thinking.

Google-reported. Results · Evaluation methods
View chart data
DeepSWE v1.1 score data
ModelReported score
Gemini 4 Argon77.9%

Vals Index

  • Gemini 4 Argon68.9%

0–100% · higher is better

An aggregate across finance, coding, legal and tax work, weighted by each sector's contribution to US GDP. Reported by Google from Vals evaluations.

Google-reported. Results · Evaluation methods
View chart data
Vals Index score data
ModelReported score
Gemini 4 Argon68.9%

AutomationBench

  • Gemini 4 Argon51.3%

0–100% · higher is better

End-to-end business workflows on Zapier's private evaluation set. Publisher-reported; this does not measure MiniRouter routes.

Google-reported. Results · Evaluation methods
View chart data
AutomationBench score data
ModelReported score
Gemini 4 Argon51.3%

Google uses its highest thinking settings and pass@1 unless specified otherwise. Results depend on the benchmark and agent setup.

Introductory pricing

Vercel AI Gateway rates per million tokens, before MiniRouter markup.

Input
$2.00
Cached input
$0.10
Output
$10.00

Google lists input/output rates of $4/$20 from December 1, 2026. Google pricing · Announcement · Gateway rates

See current MiniRouter pricing →

Try Argon

Start with a bounded output budget and select the reasoning effort your task needs.

Gemini 4 Argon · cURL
curl https://api.minirouter.sh/v1/chat/completions \
  -H "Authorization: Bearer $MINIROUTER_KEY" \
  -H 'Content-Type: application/json' \
  --data '{
    "model": "google/gemini-4-argon",
    "reasoning_effort": "high",
    "max_tokens": 8192,
    "messages": [{ "role": "user", "content": "Review this migration plan and identify its failure modes." }]
  }'

Client setup, examples and model details →

Keep exploring

Compare published results in the model rankings and coding benchmarks, or read more model releases.

For MiniRouter product updates, visit the changelog.

Sources

  1. Google Gemini 4 Argon announcement

    Published Sep 30, 2026

    Read at Google ↗
  2. Vercel Gemini 4 Argon catalog

    Published Oct 13, 2026

    Read at Vercel ↗
  3. Gemini 4 Argon evaluation methodology

    Published Sep 30, 2026

    Read at Google DeepMind ↗
  4. Google DeepMind performance table

    Published Oct 13, 2026

    Read at Google DeepMind ↗

Editorial reference revised 2026-10-13. Catalog status comes from the generated routing snapshot.

Every model, one key.