Frontier Log

Gemini 4 Carbon vs Gemini 4 Argon vs Gemini 4 Hydrogen

Gemini 4 CarbonGemini 4 ArgonGemini 4 Hydrogen
What it isGoogle's strongest model, for the hardest long-horizon problems.Google's frontier model for most coding and knowledge work.Gemini 4's fast model for everyday coding and agent tasks.
PlansGoogle AI Ultra, from $99.99 a month.Google AI Pro at $19.99 a month, and AI Ultra.Google AI Plus at $4.99 a month, AI Pro and AI Ultra.
Gemini 4 CarbonGemini 4 ArgonGemini 4 Hydrogen
Announced
Released
Input / 1M tokens$2$4 from 1 Dec 2026$2$4 from 1 Dec 2026$1.50
Cached input / 1M tokens$0.10$0.20 from 1 Dec 2026$0.10$0.20 from 1 Dec 2026$0.075
Output / 1M tokens$10$20 from 1 Dec 2026$10$20 from 1 Dec 2026$7.50
Context window1M1M1M
Output limit65K65K65K
Benchmarks
Terminal-Bench Science 0.164.8%157.6%51.4%2
OSWorld 2.1 Offline set (82 tasks), partial score81.5%374.7%477.6%5
Chartography No tools72.5%671.6%766.6%8
Agents' Last Exam v1.1 Score, all tasks (122)77.4%969.0%1071.6%11

Sources: About Gemini models - Computer - Gemini Apps Help · Gemini 4 Carbon is now available · Introducing Gemini 4 Hydrogen · Google AI Pro & Ultra — get access to Gemini 4 Argon & more

Download image

Gemini 4 CarbonGemini 4 ArgonGemini 4 Hydrogen
Other benchmarks
HLE-Diamond No toolsNo dataNo data58.8%12
Terminal-Bench 4.065.2%1357.4%No data
AutomationBench 1.0.6 Private set53.9%1451.3%15No data
AutomationBench 1.0.6 AutomationBench-AANo dataNo data76.1%16
OSWorld 2.0 (v2026.08.08) Offline set, partial scoreNo data69.2%17No data
FrontierSWE v261.2%1855.0%19No data
DeepSWE v1.1No data77.9%No data
SEC-bench Pro 26050588.2%2083.4%21No data
Agents' Last Exam v1 Pass rate, set not statedNo data39.5%22No data
GraphWalks BFS, up to 128KNo data99.7%23No data
GraphWalks BFS, 256K + 1M problems96.2%84.2%No data

Notes

  1. Terminal-Bench Science 0.1, Gemini 4 Carbon: Run by Artificial Analysis · mini-swe-agent harness · high thinking · average of 3 runs
  2. Terminal-Bench Science 0.1, Gemini 4 Hydrogen: mini-swe-agent vision harness · highest thinking · average of 5 attempts per task
  3. OSWorld 2.1, Offline set (82 tasks), partial score, Gemini 4 Carbon: Gemini CUA harness · highest thinking · max of 3 runs, one attempt per run
  4. OSWorld 2.1, Offline set (82 tasks), partial score, Gemini 4 Argon: Gemini CUA harness · highest thinking · max of 3 runs, one attempt per run
  5. OSWorld 2.1, Offline set (82 tasks), partial score, Gemini 4 Hydrogen: Gemini CUA harness · highest thinking · max of 3 runs, one attempt per run
  6. Chartography, No tools, Gemini 4 Carbon: Run by Surge AI · Surge AI's public leaderboard
  7. Chartography, No tools, Gemini 4 Argon: Run by Surge AI · Surge AI's public leaderboard
  8. Chartography, No tools, Gemini 4 Hydrogen: highest thinking
  9. Agents' Last Exam v1.1, Score, all tasks (122), Gemini 4 Carbon: ALE-Claw harness · 5-hour window · October 17, 2026 task files · safety filters on
  10. Agents' Last Exam v1.1, Score, all tasks (122), Gemini 4 Argon: ALE-Claw harness · 5-hour window · October 17, 2026 task files · safety filters on
  11. Agents' Last Exam v1.1, Score, all tasks (122), Gemini 4 Hydrogen: Antigravity-CLI (agy) version 1.4.8 harness · highest thinking · 5-hour window · October 17, 2026 task files · safety filters on
  12. HLE-Diamond, No tools, Gemini 4 Hydrogen: Run by Artificial Analysis · highest thinking
  13. Terminal-Bench 4.0, Gemini 4 Carbon: Run by Artificial Analysis · mini-swe-agent harness · high thinking · average of 3 runs
  14. AutomationBench 1.0.6, Private set, Gemini 4 Carbon: Run by Zapier · set named in the evaluation methodology the post cites
  15. AutomationBench 1.0.6, Private set, Gemini 4 Argon: Run by Zapier · set named in the evaluation methodology the post cites
  16. AutomationBench 1.0.6, AutomationBench-AA, Gemini 4 Hydrogen: Run by Artificial Analysis · highest thinking
  17. OSWorld 2.0 (v2026.08.08), Offline set, partial score, Gemini 4 Argon: Gemini CUA harness · max of 3 runs, one attempt per run
  18. FrontierSWE v2, Gemini 4 Carbon: Run by Proximal · Proximal's public leaderboard
  19. FrontierSWE v2, Gemini 4 Argon: Run by Proximal · Proximal's public leaderboard
  20. SEC-bench Pro 260505, Gemini 4 Carbon: Antigravity harness · pass@1, mean of 3 attempts · 183-task JavaScript engine split (May 2026) · LLM judge for crashes in both builds
  21. SEC-bench Pro 260505, Gemini 4 Argon: Antigravity harness · pass@1, mean of 3 attempts · 183-task JavaScript engine split (May 2026) · LLM judge for crashes in both builds
  22. Agents' Last Exam v1, Pass rate, set not stated, Gemini 4 Argon: ALE-Claw harness · 5-hour window · safety filters on
  23. GraphWalks, BFS, up to 128K, Gemini 4 Argon: 650 problems up to 128K tokens, per the model evaluation document

Related

All comparisons