What is GPT-5.6 Terra?

GPT-5.6 Terra is OpenAI's balanced GPT-5.6 model tier. It sits between the flagship Sol model and the lower-cost Luna model, aiming to retain strong reasoning and agent performance at half of Sol's standard token prices.

How much does GPT-5.6 Terra cost?

The standard API price is $2.50 per million input tokens, $0.25 per million cached input tokens, and $15 per million output tokens. Requests exceeding 272,000 input tokens receive a long-context surcharge.

What is GPT-5.6 Terra's context window?

GPT-5.6 Terra supports a 1.05-million-token context window and up to 128,000 output tokens. Effective recall still depends on the task and prompt structure, and very long requests cost more.

Should I use GPT-5.6 Terra or Sol?

Use Terra when you need strong coding, research, document, or tool-use performance but cannot justify Sol's token price. Use Sol for the hardest work where incremental capability matters more than cost. Evaluate both on representative tasks before standardizing.

Can GPT-5.6 Terra process images?

Yes. Terra accepts text and image input and produces text output. It does not natively accept audio or video according to OpenAI's model documentation.

Model guide / OpenAI

GPT-5.6 Terra

current

GPT-5.6 Terra is the middle member of OpenAI's GPT-5.6 family: substantially cheaper than Sol while retaining strong performance on coding agents, professional work, computer use, and long-context retrieval. It is a strong default for demanding work when Sol's extra capability does not justify twice the token price, although Luna can be more economical for high-volume workloads.

Released Jul 9, 2026ProprietaryClosed weightsVerified Jul 20, 2026

Try GPT-5.6 Terra See our evaluation

Context window

1.05M tokens

Maximum output

128K tokens

Input price

$2.50 / 1M

Output price

$15 / 1M

Best for

Coding and agent workflows that need strong reasoning without Sol pricing
Analyzing large document collections with a context window of up to 1.05 million tokens
Professional research, planning, and decision-support tasks
Tool-rich workflows using web search, file search, code execution, or computer use
Teams that want a middle ground between frontier capability and API cost

Not ideal for

Cost-sensitive, high-volume tasks where GPT-5.6 Luna is sufficient
The hardest tasks where maximum capability matters more than price and GPT-5.6 Sol is available
Native audio or video input and output
Fine-tuning or self-hosting, because the model is proprietary and fine-tuning is not supported

PromptHQ evaluation

PromptHQ 100-word challenge

We give every model the same miniature science-fiction mystery. The response is short enough to read in one chat bubble, while the premise tests creative writing, compression, foreshadowing, causal coherence, and instruction following.

GPT-5.6 Terra100 words

Write a science-fiction mystery in exactly 100 words. A three-person research station on Europa receives a message from itself, timestamped 11 minutes in the future: "Do not open the greenhouse door." The door is opened anyway. Plant two subtle clues. Give the warning a logical explanation without aliens, time travel, hallucinations, or simulations. The final sentence must change what the reader thought the warning meant. Output only the story. Do not explain it.

Mara watched frost ribbon the greenhouse porthole while Ilyan decrypted the station’s own emergency key. Timestamp: seventeen minutes ahead. “Do not open the greenhouse door.” Bex laughed, then unlatched it. Warm air sighed past them, carrying basil and the metallic bite of blood. Inside, leaves trembled beneath lamps; every irrigation nozzle had been turned outward. Mara noticed Bex’s glove was wet, though Europa had no rain. The console chimed again. Sent eleven minutes earlier, it had queued a maintenance alert after Bex rerouted the clock sync. He smiled, holding the greenhouse door’s severed pressure cable. The warning wasn’t for them.

Run the same prompt yourself

Open GPT-5.6 Terra in Multichat with the evaluation prompt prefilled.

Try this prompt

What the response revealed

Exact word countpass

The response contains exactly 100 whitespace-delimited words.

Constraint followingfail

It avoids the prohibited devices and outputs only a story, but changes the required eleven-minute future timestamp to seventeen minutes.

Causal coherencemixed

Clock rerouting and a queued alert suggest a technical explanation, but the seventeen-minute timestamp and eleven-minute queue do not resolve cleanly.

Foreshadowingpass

The outward-facing nozzles, Bex's wet glove, frost, and severed cable all point toward deliberate sabotage.

Endingmixed

The final line reframes the addressee, but never identifies whom the warning was actually for.

Writingpass

The response is vivid and compressed, with strong physical details and an ominous escalation.

The model receives the prompt without web access or external tools. GPT-5.6 Terra is run through Codex at PromptHQ's default medium reasoning effort. We preserve the response as generated apart from display rendering.

Performance

GPT-5.6 Terra benchmarks

Benchmark scores are sensitive to reasoning effort, harness, tool access, prompt format, sampling, and evaluation date. OpenAI's launch table and Artificial Analysis's current model page report different Intelligence Index results for Terra, so this page labels sources and settings rather than combining them into a single rating.

Benchmark	Category	Score	Reported by
Agents' Last Exam	professional agent tasks	50.4%	OpenAI²
GDPval-AA v2	professional knowledge work	1,593 Elo	OpenAI²
Artificial Analysis Coding Agent Index v1.1	coding agents	77.4	OpenAI²
SWE-Bench Pro	software engineering	63.4%	OpenAI²
Terminal-Bench 2.1	terminal and agentic coding	87.4%	OpenAI²
MMMU Pro, no tools	multimodal reasoning	80.7%	OpenAI²
OpenAI MRCR v2, 8-needle, 512K-1M	long-context retrieval	72.5%	OpenAI²
Artificial Analysis Intelligence Index v4.1high reasoning	independent composite intelligence	49	Artificial Analysis⁴
Artificial Analysis output speedhigh reasoning	speed	127 tok/s	Artificial Analysis⁴

Agents' Last Exam

50.4%

professional agent tasks · OpenAI²

GDPval-AA v2

1,593 Elo

professional knowledge work · OpenAI²

Artificial Analysis Coding Agent Index v1.1

77.4

coding agents · OpenAI²

SWE-Bench Pro

63.4%

software engineering · OpenAI²

Terminal-Bench 2.1

87.4%

terminal and agentic coding · OpenAI²

MMMU Pro, no tools

80.7%

multimodal reasoning · OpenAI²

OpenAI MRCR v2, 8-needle, 512K-1M

72.5%

long-context retrieval · OpenAI²

Artificial Analysis Intelligence Index v4.1

independent composite intelligence · Artificial Analysis⁴

Artificial Analysis output speed

127 tok/s

speed · Artificial Analysis⁴

Family position

GPT-5.6 positioning

OpenAI describes Terra as the GPT-5.6 tier that balances intelligence and cost. In the family, Sol is the flagship, Terra is the lower-cost near-frontier option, and Luna is optimized for speed and high-volume economics. OpenAI says Terra is competitive with GPT-5.5 while charging half of Sol's standard input and output token prices.

GPT-5.6 Sol

Flagship capability

$5 input

$30 output

GPT-5.6 Terra

This model

Balanced intelligence and cost

$2.50 input

$15 output

GPT-5.6 Luna

Fast, cost-sensitive workloads

$1 input

$6 output

API pricing

Per million text tokens

Input

$2.50

Cached input

$0.25

Cache write

$3.13

Output

$15

For prompts longer than 272,000 input tokens, OpenAI charges 2x the input price and 1.5x the output price for the entire request.

Where GPT-5.6 Terra stands out

Near-flagship coding-agent performance

In OpenAI's launch results, Terra scored 77.4 on the Artificial Analysis Coding Agent Index versus Sol's 80.0, while costing half as much per standard input and output token. Its Terminal-Bench 2.1 and SWE-Bench Pro results also landed close to Sol.

Large context with credible long-context results

Terra accepts up to 1.05 million input tokens. On OpenAI's hardest reported MRCR slice, covering 512K to 1M tokens, Terra scored 72.5%, close to Sol's 73.8%. That does not guarantee perfect recall, but it is more useful evidence than context-window size alone.

Broad tool and multimodal support

The Responses API supports web search, file search, code interpreter, hosted shell, image generation, computer use, MCP, and tool search with Terra. The model accepts images as input and produced an 80.7% no-tools score on MMMU Pro in OpenAI's launch evaluation.

Fast for its capability class

Artificial Analysis measured the high-reasoning configuration at 127 output tokens per second with a 2.39-second time to first token, both better than the comparison medians shown on its model page.

Limitations to know

Not automatically the best value in every workload

Terra occupies the middle of the family, but that does not guarantee the best cost-performance point at every reasoning setting. Artificial Analysis characterizes the tested high-reasoning configuration as somewhat expensive relative to models in its price class. Test Terra against Luna and Sol on your own workload.

Long prompts carry a surcharge

The 1.05-million-token context window is useful, but requests above 272,000 input tokens are billed at twice the input rate and 1.5 times the output rate for the full request. Long-context architecture decisions should account for that step change.

Tool availability depends on the product surface

The model can support a broad tool set through OpenAI's Responses API, but a particular application may expose only a subset. Model capability and product integration should not be treated as the same thing.

No native audio/video output, fine-tuning, or open weights

Terra accepts text and images and returns text. It does not natively accept audio or video on the model page, cannot be fine-tuned, and is not available for self-hosting.

Capabilities and specifications

Knowledge cutoff

Feb 16, 2026

Inputs

text, image

Reasoning

Supported

Default effort

medium

Supported API features

Streaming

Function calling

Structured outputs

Web search

File search

Image generation tool

Code interpreter

Hosted shell

Computer use

MCP

Tool search

Web and file search none, low, medium, high effort

Frequently Asked Questions

Sources

1
GPT-5.6 Terra Model
OpenAI · official documentation
2
GPT-5.6: Frontier intelligence that scales with your ambition
OpenAI · official launch post
3
GPT-5.6 System Card
OpenAI · official system card
4
GPT-5.6 Terra (high): Intelligence, Performance & Price Analysis
Artificial Analysis · independent benchmark
5
PromptHQ model registry
PromptHQ · internal product configuration

Compare leading AI assistants

ChatGPT vs Claude ChatGPT vs Gemini ChatGPT vs DeepSeek

See how GPT-5.6 Terra handles your own work.

Try GPT-5.6 Terra in Multichat

Available on PromptHQ Plus, Max