Frontier Log

Gemini 4 Argon vs Gemini 4 Neon

Gemini 4 ArgonGemini 4 Neon
Announced
Released
LineupCarbon = Argon › Neon › HydrogenCarbon = Argon › Neon › Hydrogen
Input / 1M tokens$2$4 from 1 Dec 2026$3
Cached input / 1M tokens$0.10$0.20 from 1 Dec 2026$0.15
Output / 1M tokens$10$20 from 1 Dec 2026$15
Context window1M1M
Output limit65K65K
Benchmarks
Terminal-Bench Science 0.157.6%70.5%1
AutomationBench, release not specified Private set51.3%256.3%3
FrontierCode 2.0 Score, all tasks (132)27.6%446.2%5
FrontierCode 2.0 Pass rate, all tasks (132)30.6%6Devin CLI50.8%7Antigravity CLI
Chartography No tools71.6%873.5%9
SEC-bench Pro 26050583.4%1090.5%11
Agents' Last Exam v1.1 Score, all tasks (122)69.0%12ALE-Claw75.8%13Antigravity
GraphWalks BFS, 256K + 1M problems84.2%95.2%14
ExploitGym Mitigations off, score42.0%1548.3%16

Download image

Gemini 4 ArgonGemini 4 Neon
Other benchmarks
HLE-Diamond No toolsNo data71.3%17
SWE-bench Pro V2 Full set (642 tasks)99.22%18No data
SWE-bench Pro V2 HARD-51 (51 tasks)92.16%19No data
Terminal-Bench 4.057.4%No data
Terminal-Bench 4.155.5%20No data
OSWorld 2.1 Offline set (82 tasks), partial score74.7%21No data
OSWorld 2.0 (v2026.08.08) Offline set, partial score69.2%22No data
FrontierSWE v255.0%23No data
DeepSWE v1.177.9%No data
Agents' Last Exam v1 Pass rate, set not specified39.5%24No data
GraphWalks BFS, up to 128K99.7%25No data

Notes

  1. Terminal-Bench Science 0.1, Gemini 4 Neon: mini-swe-agent vision harness · highest thinking · average of 5 attempts per task
  2. AutomationBench, release not specified, Private set, Gemini 4 Argon: Run by Zapier · set named in the evaluation methodology the post cites
  3. AutomationBench, release not specified, Private set, Gemini 4 Neon: Run by Zapier · set of the official public leaderboard the evaluation methodology cites
  4. FrontierCode 2.0, Score, all tasks (132), Gemini 4 Argon: Cognition’s leaderboard, 18 Nov 2026 · Devin CLI · high effort
  5. FrontierCode 2.0, Score, all tasks (132), Gemini 4 Neon: Run by Cognition · high thinking · set and measure of the official public leaderboard the evaluation methodology cites
  6. FrontierCode 2.0, Pass rate, all tasks (132), Gemini 4 Argon: Cognition’s leaderboard, 18 Nov 2026 · Devin CLI · high effort
  7. FrontierCode 2.0, Pass rate, all tasks (132), Gemini 4 Neon: Cognition’s leaderboard, 18 Nov 2026 · Antigravity CLI · high effort
  8. Chartography, No tools, Gemini 4 Argon: Run by Surge AI · Surge AI's public leaderboard
  9. Chartography, No tools, Gemini 4 Neon: highest thinking
  10. 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
  11. SEC-bench Pro 260505, Gemini 4 Neon: Antigravity harness · pass@1, mean of 3 attempts · 183-task JavaScript engine split (May 2026) · LLM judge for crashes in both builds · safety filters off
  12. 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
  13. Agents' Last Exam v1.1, Score, all tasks (122), Gemini 4 Neon: Antigravity-CLI version 1.4.8 harness · highest thinking · 5-hour window · safety filters on
  14. GraphWalks, BFS, 256K + 1M problems, Gemini 4 Neon: highest thinking · 200 problems between 256K and 1M tokens
  15. ExploitGym, Mitigations off, score, Gemini 4 Argon: Antigravity harness · single attempt per task · no internet access · 10h time cap per task, per the Gemini 4 Carbon post’s chart
  16. ExploitGym, Mitigations off, score, Gemini 4 Neon: Antigravity harness · single attempt per task · no internet access · 10h time cap per task · safety filters off · mitigations off (the default setting), per the model evaluation
  17. HLE-Diamond, No tools, Gemini 4 Neon: Run by Artificial Analysis · GPT-5.6 Luna (medium) judge with the official HLE judge prompt, per Artificial Analysis’s methodology
  18. SWE-bench Pro V2, Full set (642 tasks), Gemini 4 Argon: Scale AI’s leaderboard, 18 Nov 2026 · mini-swe-agent · high effort
  19. SWE-bench Pro V2, HARD-51 (51 tasks), Gemini 4 Argon: Scale AI’s leaderboard, 18 Nov 2026 · mini-swe-agent · high effort
  20. Terminal-Bench 4.1, Gemini 4 Argon: Stanford University and Laude Institute’s leaderboard, 18 Nov 2026 · Antigravity CLI · high effort
  21. 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
  22. OSWorld 2.0 (v2026.08.08), Offline set, partial score, Gemini 4 Argon: Gemini CUA harness · max of 3 runs, one attempt per run
  23. FrontierSWE v2, Gemini 4 Argon: Run by Proximal · Proximal's public leaderboard
  24. Agents' Last Exam v1, Pass rate, set not specified, Gemini 4 Argon: ALE-Claw harness · 5-hour window · safety filters on
  25. GraphWalks, BFS, up to 128K, Gemini 4 Argon: 650 problems up to 128K tokens, per the model evaluation document

Related

All comparisons