SYLIN

GPT Terra

by OpenAI · Model · Language & code models

unproven

OpenAI's mid-priced model; claims of matching Sol are unconfirmed by outsiders

OpenAI's mid-tier model between Luna and Sol, built for computer use; on all paid plans in ChatGPT, Codex and API

Why we say so

OpenAI says Terra 6.1 matches Sol on professional tasks at lower cost and trails Claude Sonnet 5.5 on Terminal-Bench. Google's table shows its Gemini Hydrogen ahead of Terra, but that is also a maker's claim. No independent benchmark or hands-on review is on record.

Versions

VersionReleasedRank
GPT-6.1 TerraLATEST
unproven Only OpenAI's claims so far
Nov 1
GPT-5.6 Terra

Evidence

  1. Nov 1
    Matches larger Sol on the Agent/Exam professional-tasks benchmark at much lower cost; lower factual error rate
  2. Nov 1
    Cheaper than Claude Sonnet 5.5 but scores lower on Terminal-Bench; roughly 10x cheaper per the video
  3. Aug 28
    Listed at $2.00 input, $0.20 cached input, $2.50 cache writes, $12.00 output per 1M tokens
  4. Aug 22
    GPT Terra (listed as GPT-5.6 Terra) scored 3,214 (11.0% gap closed), below Sol and Luna

Cost to run

Alternatives

Claude OpusTop of Terminal-Bench-Science at 30.0% resolution with Claude Code, ahead of GPT Sol 22.4% and Fable 21.4%
Claude FableFable 5 (Claude Code, high effort) topped the speedrun: record 2,726, 81.7% of the gap to human record closed, 8.7 agent-days