GPT 5.6 Terra

What is GPT 5.6 Terra?#

GPT 5.6 Terra is OpenAI's everyday model in the GPT 5.6 family, previewed on June 26, 2026 and generally available since July 9, 2026. It runs on Unifically as openai/gpt-5.6-terra. It takes text and image input and returns text, with a 1,050,000-token context window, a 128,000-token max output, and knowledge up to February 16, 2026. The pitch is simple: output competitive with GPT-5.5 across most workloads, and ahead of it on agent benchmarks. It beats Claude Fable 5 on Agents' Last Exam 50.4 to 40.5 and edges it on the Coding Agent Index. Sol sits above it for the hardest problems and Luna below it for the fastest, high-volume calls.

Key features of GPT 5.6 Terra#

Beats Fable 5 on Agents' Last Exam

Terra scores 50.4 on Agents' Last Exam, which tests long-running professional workflows across 55 fields. That is ahead of GPT-5.5 at 46.9, Claude Opus 4.8 at 45.2, and Claude Fable 5 at 40.5.

Edges Fable 5 on the Coding Agent Index

Terra scores 77.4 on the Artificial Analysis Coding Agent Index v1.1, just above Claude Fable 5 at 77.2 and GPT-5.5 at 76.4, in roughly a third of the time with about half the output tokens. On Terminal-Bench 2.1 it reaches 84.3% at max effort.

Reasoning effort you can tune

Terra runs from no reasoning up to max effort, so token spend is a dial you control per task. On GeneBench v1 that range takes it from 2.2% to 28.4% pass@1: the same model serves quick replies and long analyses.

Stronger security work than GPT-5.5

At max effort Terra reaches 52.9% on ExploitBench and 57.7% on SEC-Bench Pro, versus 47.9% and 45.8% for GPT-5.5, plus 91.8% on OpenAI's capture-the-flag suite. Useful for supervised code review and vulnerability triage.

Beats Opus 4.8 on CursorBench 3.2

On Cursor's coding-agent benchmark, Terra scores 64.9% at max effort, ahead of Opus 4.8 at 62.3% and Sonnet 5 at 61.5%. Real Cursor edits, not just terminal tasks.

Programmatic tool calling

With Programmatic Tool Calling in the Responses API, Terra writes and runs small programs that coordinate tools, filter intermediate results, and keep only what matters. Tool-heavy tasks finish with fewer round trips and fewer tokens.

Best for#

Use cases#

Swap Terra in wherever GPT-5.5 runs today: coding agents, support assistants, document Q&A, and tool-calling backends keep their quality. Build an agent loop that runs at low effort for routine turns and jumps to max effort when a task gets hard, since the same model covers both ends. Terra reads screenshots and charts inline, so it fits vision-assisted flows like reviewing a UI mockup or reading a scanned invoice. It also fits supervised code-review pipelines, where it beats GPT-5.5 on ExploitBench at every effort level above medium.

Limitations#

Terra trails Sol where reasoning depth decides the outcome: 84.3% versus 88.8% on Terminal-Bench 2.1, 52.9% versus 73.5% on ExploitBench, and it has no ultra mode. For the hardest agent runs, Sol is worth the step up.

GPT 5.6 runs behind layered safeguards. Real-time cyber and biology classifiers can pause or withhold higher-risk generations, and some legitimate dual-use security work may be blocked or slowed.

Arena has not published an Elo for Terra yet. A ranked position is pending while the public rollout is this new.

GPT 5.6 Terra vs GPT-5.5#

Terra beats GPT-5.5 on most of the published agent evaluations: 50.4 to 46.9 on Agents' Last Exam, 77.4 to 76.4 on the Coding Agent Index, 69.6% to 67.0% on DeepSWE v1.1, 91.8% to 88.1% on capture-the-flag, and 52.9% to 47.9% on ExploitBench. GPT-5.5 keeps a small lead on Terminal-Bench 2.1, 85.6% to 84.3%, and at mid efforts on GeneBench v1. For most workloads Terra is the newer, more efficient pick; check terminal-heavy pipelines against your own tasks first.

When to use GPT 5.6 Terra#

Terra is the default of the GPT 5.6 family. Use it for the everyday work that used to go to GPT-5.5: coding, assistants, extraction, and tool-heavy backends. Move up to Sol when a task needs max-effort depth or ultra mode's parallel subagents, and down to Luna when volume and latency matter more than reasoning score.

API examples#

Call GPT 5.6 Terra from any language by POSTing to /v1/chat/completions, the OpenAI-compatible endpoint shared by every language model on the platform. Full parameter docs live at docs.unifically.com/models/llm/openai/gpt-5.6-terra.

curl -X POST https://api.unifically.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "openai/gpt-5.6-terra",
    "messages": [
      { "role": "user", "content": "Summarize this support thread and draft a reply." }
    ]
  }'

The response comes back synchronously with the completion. Set "stream": true to receive tokens as they generate.