Frontier Log

GPT-6 Astra vs GPT-6.1 Sol vs GPT-6.1 Terra

GPT-6 AstraGPT-6.1 SolGPT-6.1 Terra
What it isOpenAI's most capable model, for the most demanding work.An upgrade to GPT-6 Sol with near-Astra performance for complex work at a lower cost.A fast, efficient model for visual reasoning, computer use and writing.
PlansPro in Chat, Work and Codex, and Plus in Work and Codex.Plus and Pro, in Work and Codex.Plus and Pro, in Work and Codex.
GPT-6 AstraGPT-6.1 SolGPT-6.1 Terra
Released
Input / 1M tokens$10$2$1
Cached input / 1M tokens$1$0.10$0.05
Output / 1M tokens$50$10$5
Context window1.05M1.05M1.05M
Output limit128K128K128K
Benchmarks
Terminal-Bench 4.057.9%155.9%250.3%3
Terminal-Bench Science 0.164.6%457.0%558.3%6
OSWorld 2.1 Offline set (82 tasks), scoring not stated80.9%778.0%874.2%9
MMMU-Pro Vision setting86.7%1086.3%1185.5%12

Sources: GPT-6 Astra · Introducing GPT-6.1 Sol · GPT-6.1 Sol · GPT-6.1 Terra · help.openai.com · Introducing GPT-6.1 Terra

Download image

GPT-6 AstraGPT-6.1 SolGPT-6.1 Terra
Other benchmarks
Humanity's Last Exam With tools, set not stated57.2%13No dataNo data
GPQA Diamond96.0%No dataNo data
BrowseComp91.5%No dataNo data
ARC-AGI-295.0%No dataNo data
AutomationBench 1.0.6 Private set41.4%1436.1%15No data
OSWorld 2.0 (v2026.08.08) Offline set, partial score72.6%71.4%16No data
FrontierCode 1.1 Main set53.3%17No dataNo data
FrontierCode 1.1 Extended set64.5%18No dataNo data
DeepSWE v1.174.1%75.2%19No data
ARC-AGI-1 Semi-Private set (100 tasks)98.5%No dataNo data
ARC-AGI-1 Public set (400 tasks)98.5%No dataNo data
ARC-AGI-3 Semi-Private set (55 environments)99.9%20No dataNo data
SEC-bench Pro, release not stated85.4%No dataNo data
Agents' Last Exam v1 Score, all tasks (152)59.3%21No dataNo data
Agents' Last Exam v1 Measure and set not statedNo data58.6%2260.0%23

Notes

  1. Terminal-Bench 4.0, GPT-6 Astra: high effort
  2. Terminal-Bench 4.0, GPT-6.1 Sol: max effort
  3. Terminal-Bench 4.0, GPT-6.1 Terra: maximum effort
  4. Terminal-Bench Science 0.1, GPT-6 Astra: max effort
  5. Terminal-Bench Science 0.1, GPT-6.1 Sol: max effort
  6. Terminal-Bench Science 0.1, GPT-6.1 Terra: max effort
  7. OSWorld 2.1, Offline set (82 tasks), scoring not stated, GPT-6 Astra: max effort
  8. OSWorld 2.1, Offline set (82 tasks), scoring not stated, GPT-6.1 Sol: max effort
  9. OSWorld 2.1, Offline set (82 tasks), scoring not stated, GPT-6.1 Terra: max effort
  10. MMMU-Pro, Vision setting, GPT-6 Astra: max effort
  11. MMMU-Pro, Vision setting, GPT-6.1 Sol: max effort
  12. MMMU-Pro, Vision setting, GPT-6.1 Terra: max effort
  13. Humanity's Last Exam, With tools, set not stated, GPT-6 Astra: maximum at any effort · run in OpenAI's research environment or via its API
  14. AutomationBench 1.0.6, Private set, GPT-6 Astra: Run by Zapier · highest score at any effort · set named in the GPT-6.1 Sol post, which prints the same figure
  15. AutomationBench 1.0.6, Private set, GPT-6.1 Sol: Run by Zapier · max effort
  16. OSWorld 2.0 (v2026.08.08), Offline set, partial score, GPT-6.1 Sol: max effort
  17. FrontierCode 1.1, Main set, GPT-6 Astra: run with a developer message similar to part of its developer message in Codex
  18. FrontierCode 1.1, Extended set, GPT-6 Astra: run with a developer message similar to part of its developer message in Codex
  19. DeepSWE v1.1, GPT-6.1 Sol: high effort
  20. ARC-AGI-3, Semi-Private set (55 environments), GPT-6 Astra: OpenAI's Responses API harness, which changes two settings (retained reasoning and compaction, per OpenAI's ARC-AGI-3 harness post)
  21. Agents' Last Exam v1, Score, all tasks (152), GPT-6 Astra: harness not stated · best of any effort
  22. Agents' Last Exam v1, Measure and set not stated, GPT-6.1 Sol: harness not stated · max effort
  23. Agents' Last Exam v1, Measure and set not stated, GPT-6.1 Terra: harness not stated · max effort

Related

All comparisons