Frontier Log

Gemini 4 Argon vs GPT-6 Sol

Gemini 4 ArgonGPT-6 Sol
Announced
Released
Input / 1M tokens$2$4 from 1 Dec 2026$2
Cached input / 1M tokens$0.10$0.20 from 1 Dec 2026$0.20
Output / 1M tokens$10$20 from 1 Dec 2026$10
Context window1M1.05M
Output limit65K128K
Benchmarks
AutomationBench 1.0.6 Private set51.3%133.2%2
OSWorld 2.0 (v2026.08.08) Offline set, partial score69.2%360.5%4
DeepSWE v1.177.9%68.8%5

Download image

Gemini 4 ArgonGPT-6 Sol
Other benchmarks
Terminal-Bench 4.057.4%No data
Terminal-Bench Science 0.157.6%No data
OSWorld 2.1 Offline set (82 tasks), partial score74.7%6No data
FrontierSWE v255.0%7No data
Chartography No tools71.6%8No data
SEC-bench Pro 26050583.4%9No data
Agents' Last Exam v1 Pass rate, set not stated39.5%10No data
Agents' Last Exam v1 Measure and set not statedNo data56.4%11
Agents' Last Exam v1.1 Score, all tasks (122)69.0%12No data
GraphWalks BFS, up to 128K99.7%13No data
GraphWalks BFS, 256K + 1M problems84.2%No data

Notes

  1. AutomationBench 1.0.6, Private set, Gemini 4 Argon: Run by Zapier · set named in the evaluation methodology the post cites
  2. AutomationBench 1.0.6, Private set, GPT-6 Sol: Run by Zapier · xhigh effort
  3. OSWorld 2.0 (v2026.08.08), Offline set, partial score, Gemini 4 Argon: Gemini CUA harness · max of 3 runs, one attempt per run
  4. OSWorld 2.0 (v2026.08.08), Offline set, partial score, GPT-6 Sol: xhigh effort
  5. DeepSWE v1.1, GPT-6 Sol: max effort
  6. 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
  7. FrontierSWE v2, Gemini 4 Argon: Run by Proximal · Proximal's public leaderboard
  8. Chartography, No tools, Gemini 4 Argon: Run by Surge AI · Surge AI's public leaderboard
  9. 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
  10. Agents' Last Exam v1, Pass rate, set not stated, Gemini 4 Argon: ALE-Claw harness · 5-hour window · safety filters on
  11. Agents' Last Exam v1, Measure and set not stated, GPT-6 Sol: harness not stated · max effort
  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. GraphWalks, BFS, up to 128K, Gemini 4 Argon: 650 problems up to 128K tokens, per the model evaluation document

Related

All comparisons