Changelog

Gemini 4 Carbon is here.

Google’s strongest Gemini 4 model, at Argon’s price per token.

How it scores.

Artificial Analysis Intelligence Index

  • Fable 5.264.6
  • Carbon high60
  • Opus 5.559.1
  • Carbon medium58.3
  • GPT-6 Astra56.9
  • GPT-6.1 Sol54.6
  • Carbon low54.2
  • Argon high50.4

0–70 · higher is better

Artificial Analysis, Index v5, read 10 November 2026. Claude and GPT models at their best effort, each on its publisher’s API.
View chart data
Artificial Analysis Intelligence Index scores
ModelIndex
Fable 5.264.6
Carbon high60.0
Opus 5.559.1
Carbon medium58.3
GPT-6 Astra56.9
GPT-6.1 Sol54.6
Carbon low54.2
Argon high50.4

Where it pulls ahead of Argon.

Carbon is built on Argon and spends more thinking on each turn. Google still points most work at Argon.

GraphWalks · 256K to 1M tokens

  • Carbon96.2%
  • Argon84.2%

Terminal-Bench-Science 0.1

  • Carbon64.8%
  • Argon52.9%

Agents' Last Exam v1.1

  • Carbon77.4%
  • Argon69.0%

AutomationBench

  • Carbon53.9%
  • Argon51.3%
Google, results as of October 2026. 0–100%, publisher-reported at the highest thinking setting. GPT-6 Astra leads FrontierSWE 2 and Claude Opus 5.5 leads OSWorld-2.1. Full table.
View chart data
Gemini 4 Carbon and Gemini 4 Argon launch scores
BenchmarkCarbonArgon
GraphWalks · 256K to 1M tokens96.2%84.2%
Terminal-Bench-Science 0.164.8%52.9%
Agents' Last Exam v1.177.4%69%
AutomationBench53.9%51.3%

Pick an effort, copy the request.

Medium is the default. Set MINIROUTER_KEY to your API key.

Intelligence Index58.3
First answer token4.9 sInside the 95-second limit for a streamed reply to start.
Gemini 4 Carbon · medium · cURL
curl https://api.minirouter.sh/v1/chat/completions \
  -H "Authorization: Bearer $MINIROUTER_KEY" \
  -H 'Content-Type: application/json' \
  --data '{
  "model": "google/gemini-4-carbon",
  "reasoning_effort": "medium",
  "max_tokens": 32768,
  "messages": [
    {
      "role": "user",
      "content": "Review this diff for bugs and missing tests."
    }
  ]
}'

Index and wait times: Artificial Analysis, measured on Google’s API and read 10 November 2026. The line marks 95 seconds.

Before you switch.

  • max_tokens counts thinking. Google recommends at least 32,768.
  • Google ignores temperature, top_p and top_k.
  • A streamed reply must start within 95 seconds, so high effort can time out.
  • Google’s list price doubles on 1 December 2026, for Argon too.

See what else shipped.