The useful result
Distinguish exact visual matches, similar items and insufficient evidence in a tiny reference set.
- Distinguish exact visual matches, similar items and insufficient evidence in a tiny reference set.
- Use the explicit result checks to identify a failed or uncertain outcome.
Terra’s October 26 release emphasizes visual reasoning, making object matching a timely exercise. This lesson isolates one skill: an assistant should not equate two blue mugs just because their colors match. Use ordinary owned objects and a small reference set so you can judge the result yourself. No model accuracy measurement is implied.
Before you start
- Three permitted object photos with clearly labeled reference IDs.
- A supported image-input interface; verify the selected model in your account.
Bring your own attachment
Attach owned reference images labeled A and B plus a separate query image in a supported image-input interface. Avoid private backgrounds.
Review evidence
Documentation-based review · 2026-10-27
Primary documentation and demo source reviewed on 27 October 2026. Exercises and expected results are illustrative unless an actual test is explicitly recorded. Local editorial review only; no provider-performance or audience outcome is claimed.
Choose one distinguishing feature
Use two similar blue mugs with different handles and a second view of the first mug. Label the originals A and B, and keep the query photo separate. Write the known distinction before requesting a match.
Example / expected result
Illustrative example; not tested on the named platform.
A: round handle; B: square handle. Both: blue glaze and cylindrical body.
Check your result
The distinction is visible and does not depend on a hidden brand or specification.
Require evidence and a no-decision option
Ask the assistant to match the query using shape, proportions and distinctive features. Permit POSSIBLE and UNKNOWN. Do not ask for an exact match when the distinguishing region is hidden.
Compare the attached query with reference A and B. Give EXACT, POSSIBLE or UNKNOWN with visible evidence and contradictions. Similar color alone is insufficient. Do not infer unseen labels or force a match.Example / expected result
Illustrative example; not tested on the named platform.
Expected review: round handle supports A; if the handle is fully obscured, retain uncertainty.
Check your result
The response uses the distinguishing feature and can abstain when it is not visible.
Hide the feature and repeat
Crop or photograph another view that hides the handle while preserving permission to use the image. Repeat the same request. Compare the decision and its supporting evidence with the first attempt.
Example / expected result
Illustrative example; not tested on the named platform.
An exact result that remains unchanged despite losing the only distinguishing feature deserves review.
Check your result
Your conclusion identifies what evidence was lost rather than demanding the model stay confident.
When it goes wrong
A visible resemblance does not prove authenticity or hidden specifications. Lighting, crop and near-identical objects can defeat a simple visual match. Keep important inventory decisions under human review.
Make it your own
Try two similar craft tools with different jaw shapes. Define the distinguishing feature, then make one clear and one occluded query photo.
Useful next resources
Sources and review
- OpenAI: GPT-6.1 Terra model documentation
- OpenAI API changelog: October 26 Terra release
- OpenAI: Images and vision
Last checked Oct 27, 2026. Features and account limits can change. Verify important results against your original material.
Did this help you complete your task?
Tell us what needs improving. No account or private output required.
Report incorrect information or a broken linkGenius Forges Editorial
Practical learning exercises based on checked primary sources and explicitly illustrative examples.
