GPT 5.6 Terra on VM0. The balanced GPT-5.6 workhorse
The balanced middle tier of OpenAI's GPT-5.6 preview family. Most of Sol's agentic quality at half the credit cost — the sensible everyday default on the OpenAI side.
400K tokens · Text / Vision / Code · Prompt cache
GPT 5.6 Terra is the balanced middle tier of OpenAI's GPT-5.6 preview generation — the everyday workhorse that handles the bulk of agentic coding and tool-use work without the flagship price. It keeps most of Sol's behavioural gains over GPT-5.5 while billing at half the rate.
Vendor list price is $2.50 / $15 per 1M tokens with cached input at $0.25 / 1M. It sits at the ×1 credit tier on VM0 — the balanced band — which makes it the natural everywhere-default in a GPT-5.6 agent, with Sol reserved for escalation and Luna for high-volume or latency-sensitive steps. GPT-5.6 is currently a preview family; GPT-5.5 remains the default OpenAI model until it graduates.
What is GPT 5.6 Terra?
July 2026 preview (middle tier of the GPT-5.6 family) · Balanced middle tier of the GPT-5.6 family, between flagship Sol and economy Luna. The recommended everyday default for GPT-5.6 agents.
GPT-5.6 arrived in July 2026 as a three-tier family — Sol, Terra and Luna. Terra is the middle tier: OpenAI positions it as the balanced default that covers most agentic-coding and tool-use work, mirroring the role GPT-5.4 played in the previous generation. It shares the 400K-token context window and the reasoning_effort parameter with the rest of the family, so it drops into existing Codex agents unchanged.
Against Sol, Terra trades a slice of top-end reasoning depth for half the list price ($2.50 / $15 per 1M) and a ×1 credit multiplier on VM0. In OpenAI's preview materials it stays close to Sol on routine coding and tool-routing tasks and only falls behind on the hardest multi-file patches, long orchestration loops and graduate-level reasoning — which is exactly the boundary at which you escalate to Sol.
As with the rest of the preview family, published numbers are early and will move, and OpenAI has flagged SWE-bench Verified contamination across frontier models. The durable signal is that Terra gives you most of the family's tool-call accuracy and first-attempt patch quality at the balanced price band — the reason it, not Sol, is meant to run everywhere.
What's notable about GPT 5.6 Terra
Headline architecture and capability features.
GPT 5.6 Terra keeps the 400K-token context window, billed at standard input pricing across the entire window. It supports the reasoning_effort parameter, prompt caching where cached input bills at one-tenth the input rate ($0.25 / 1M) plus a cache-write charge ($3.125 / 1M), and the Responses API surface the codex CLI uses by default. Tool-use, structured outputs and computer-use match the rest of the GPT-5.6 family. Inputs are multimodal across text, vision and code; there is no native image generation.
Specs at a glance
GPT 5.6 Terra benchmarks
Preview figures from OpenAI's GPT-5.6 materials, shown against the public GPT-5.4 numbers Terra succeeds. Treat all percentages as directional: the family is in preview and OpenAI has flagged SWE-bench Verified contamination across frontier models.
GPT 5.6 Terra pricing
Provider list price, per 1M tokens.
How GPT 5.6 Terra behaves in practice
Observed behaviour from production agent runs.
Tool routing
Close to Sol on routine and moderately hard tool-routing. The gap opens only on the hardest edge cases — conditional selection and tool calls after long reasoning — where Sol's extra depth pays off.
First-attempt code edits
Strong patch quality on single- and few-file changes. Reach past Terra to Sol when a patch spans many files and must apply cleanly the first time; for most edits Terra lands them without a wasted CI run.
Computer use
Reliable on short-to-medium GUI sequences. For long multi-step computer-use runs where a mid-session derailment is expensive, Sol's higher OSWorld score is worth the premium.
Speed
Faster than Sol and a comfortable middle ground — around 90 tokens/sec at medium effort in early testing. Fast enough for interactive agents while keeping real reasoning depth.
Cost profile
Half of Sol's list price and the ×1 credit tier make Terra the model you leave running everywhere. The economics only tip toward Luna on very high-volume or latency-critical steps.
Best agent tasks for GPT 5.6 Terra
The everyday coding agent
Use Terra as the default for the day-to-day work of a coding agent: reading code, writing functions, running tests, applying single- and few-file patches. It clears the bar on most tasks and keeps the credit bill at ×1.
The sub-agent under a Sol orchestrator
When Sol plans a ten-step job, Terra is the tier that executes most of those steps. You get near-flagship quality on the execution layer while paying the flagship rate only at the planner.
Interactive tool-use sessions
For an agent that a human is watching in real time, Terra's faster generation keeps the loop responsive while still handling conditional tool selection and structured outputs reliably.
Mixed workloads that don't justify the flagship
Support triage, doc drafting, code review comments, moderate refactors — the wide band of real work that needs solid reasoning but not the absolute frontier. Terra covers it at half the cost of Sol.
When to skip GPT 5.6 Terra
Skip Terra on the hardest multi-file refactors, long orchestration loops and graduate-level reasoning where Sol visibly does better, and on high-volume bulk classification or latency-critical replies where GPT 5.6 Luna is cheaper and faster.
GPT 5.6 Terra vs other models
GPT 5.6 Terra vs GPT 5.6 Sol
Sol is the flagship you escalate to; Terra is the default you run everywhere. Terra keeps most of Sol's tool-routing and coding quality at half the list price and half the credit cost — promote only when a step visibly needs Sol's extra reasoning depth.
GPT 5.6 Terra vs GPT-5.5
Terra is the balanced GPT-5.6 tier, priced below the previous flagship GPT-5.5 while, in preview, matching or beating it on routine agentic work. Where GPT-5.5 was the escalation tier of its generation, Terra is meant to be the everyday default of the new one.
GPT 5.6 Terra vs Claude Sonnet 5
Peer balanced workhorses in different families. Sonnet 5 brings the 1M-token context window and Anthropic's ecosystem; Terra brings the Codex framework and OpenAI's computer-use profile. Pick by which framework your existing agents target and which context length you need.
Bottom line: should you use GPT 5.6 Terra?
GPT 5.6 Terra is the sensible everyday default of the GPT-5.6 family: most of Sol's quality at half the cost. Run it everywhere, escalate to Sol for the hardest steps, drop to Luna for bulk and latency.
Frequently asked questions
What is GPT 5.6 Terra's context window?
400,000 tokens, with up to 128K tokens of output per response. The full window bills at standard rates.
How is Terra different from Sol?
Terra is the balanced middle tier: it keeps most of Sol's agentic-coding and tool-use quality at half the list price ($2.50 / $15 vs $5 / $30) and half the VM0 credit cost. Sol pulls ahead only on the hardest multi-file patches, long orchestration loops and graduate-level reasoning.
Is GPT-5.6 generally available on VM0?
It is in preview. All three tiers are selectable, but GPT-5.5 remains the default OpenAI model until the GPT-5.6 family graduates, and preview benchmark numbers are early.
Does GPT 5.6 Terra support prompt caching?
Yes. Cached input bills at $0.25 per 1M tokens with a cache-write charge of $3.125 per 1M. Worth enabling whenever your system prompt or tool schema is stable across calls.
What framework does GPT 5.6 Terra use on VM0?
Codex. VM0 routes GPT-5.6 through the Codex framework's Responses API surface. Claude Code-framework agents are not compatible with GPT-5 models on VM0.
Alternatives
Using GPT 5.6 Terra on VM0
Two ways to access GPT 5.6 Terra on VM0
VM0 supports GPT 5.6 Terra as a Built-in model billed in VM0 credits, and through bring-your-own with a OpenAI API key. The Built-in path uses VM0 Managed routing and the credit multiplier explained below; the bring-your-own path bills you directly with the upstream vendor and skips the VM0 credit conversion entirely.
VM0's recommendation
VM0 positions GPT 5.6 Terra as a core agent model, recommended alongside Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6 for the steps that drive the actual outcome of an agent run. These are the models we'd pick for the orchestrator role, for code-touching agents, and for any step where a wrong answer is expensive.
Credits and the ×1 multiplier
Every Built-in model on VM0 is priced as a multiple of Claude Sonnet 4.6, which sits at the ×1 credit baseline. GPT 5.6 Terra bills at ×1 credits. The multiplier is what shows up on your VM0 invoice; the vendor list price in the pricing table above is what the upstream provider charges before VM0 converts it into credits.
GPT 5.6 Terra sits at the ×1 baseline that every other Built-in model is priced against, so it's the unit you compare costs in when picking between models on VM0.
Available on VM0 since July 2026 (preview).