AI Agent Hub
Back to models
OpenAI: GPT Terra Latest logo

OpenAI: GPT Terra Latest

Closed Source ~openai Released 2026-09-11
42.0 / 100 1.1M context Proprietary

About this model

OpenAI: GPT Terra Latest is a rolling API alias that resolves to the newest model in the GPT Terra tier. As of its introduction, that target is GPT-5.6 Terra, the mid-tier member of the GPT-5.6 family launched in July 2026 alongside flagship Sol and budget Luna. Terra is positioned for everyday professional work where teams want strong reasoning, tool use, and multimodal understanding without Sol-level cost. It supports a very large context window (about 1.05M tokens), up to 128K output tokens per response, configurable reasoning effort from none through max, programmatic tool calling, function calling, and native multimodal inputs including text and images.

On OpenAI’s GPT-5.6 evaluation suite, Terra delivers competitive frontier-class results at substantially lower estimated cost than GPT-5.6 Sol. Highlights include 92.9% on GPQA Diamond, 87.4% on Terminal-Bench 2.1, 63.4% on SWE-Bench Pro, 69.6% on DeepSWE, 87.5% on BrowseComp, 50.4% on Agents’ Last Exam, and 80.7–82% on MMMU-Pro depending on tool access. Independent trackers also report strong coding-agent results, including roughly 95% on SWE-bench Verified (Vals AI). Terra is described as outperforming prior GPT-5.5 peak performance on several workloads while using fewer tokens and less latency than larger tiers.

GPT Terra Latest is proprietary and available only through hosted APIs (OpenAI API, ChatGPT Work, Codex, and partner routes such as Azure OpenAI and OpenRouter). Pricing for the underlying Terra snapshot is typically about $2 per million input tokens and $12 per million output tokens, with prompt caching discounts. Parameter count and training compute are not disclosed. Knowledge cutoff for the current Terra generation is February 16, 2026.

Benchmark Scores

ARC
84.0
DeepSWE
69.6
MMMU-Pro
82.0
SimpleQA
43.0
AIME-2024
100.0
AIME-2025
100.0
Arena-Elo
1505.0
GDPval-AA
1593.0
BrowseComp
87.5
GPQA-Diamond
92.9
SWE-Bench-Pro
63.4
AutomationBench
15.2
Agents-Last-Exam
50.4
SWE-Bench-Verified
95.4
Terminal-Bench-2.1
87.4
Toolathlon-Verified
53.1

Technical Specs

  • Architecture: Transformer
  • Context Window: 1,050,000 tokens
  • Input Modalities: text, image

Hardware Requirements

  • Compute: API only

Pricing

Input Output Currency
2.00 / 1M tokens 12.00 / 1M tokens USD