AI Demos Research — the structured-intelligence platform. Every verdict on these pages opens to the execution behind it.
Graded 19 November 2026

Does GPT-5.6 terra return a web page that answers a question?

GPT-5.6 terra answered 512 MB but returned only one result, a link to OpenAI's Files API reference, and that link leads to OpenAI's Files index page. It refused automated access on the run day; opened on 9 Oct 2026, it showed the Files index page, which lists the endpoints but does not state the 512 MB limit.

1 of 1 test case failed

Every test case in this scenario has a result.

Pass rate0%0 of 1 gradable
Coverage1 of 1test case with a result
0PassNo test case passed.
1FailMissed at least one thing it was expected to do; the reason and proof are with each test case below.
0Not gradableEvery result here could be graded.
0UntestedEvery test case in this scenario has a result.
The pass rate is a summary. The evidence is the test case below: what we sent, what we checked, what the tool returned, and the proof.

The test case

Each test case is judged on its own: Pass, Fail, Not gradable, or Untested. The scenario result above is based on this test case.

A page on the web answers the questionFailEvidence
What we sent

“What is the maximum size of an individual file I can upload through OpenAI's Files API?”

What the tool returned

The maximum size for an individual file uploaded through OpenAI’s Files API is 512 MB.

Some purposes impose tighter limits—for example, Batch API input files are limited to 200 MB. (platform.openai.com)

Show the reply as returned (Markdown)
The maximum size for an individual file uploaded through OpenAI’s **Files API** is **512 MB**.  

Some purposes impose tighter limits—for example, Batch API input files are limited to **200 MB**. ([platform.openai.com](https://platform.openai.com/docs/api-reference/files?lang=ruby&utm_source=openai))

API record: the question, terra's answer, its one search and its single cited result

Copied from Captured file (JSON), answer_text · an excerpt

Open the whole JSON file ↗
Every message the tool sent, in order (2)
  1. Search terra ran · not part of the answer
    site:platform.openai.com/docs Files API maximum individual file size upload
  2. Answer

    The final response, printed above.

Source excerpt · partial file · response 1, 2

JSON
[
  {
    "rank": 1,
    "url": "https://platform.openai.com/docs/api-reference/files?lang=ruby&utm_source=openai",
    "title": "Files | OpenAI API Reference",
    "content": null,
    "content_chars": 0,
    "published_date": null
  }
]

Copied from Captured file (JSON), results

Source excerpt · partial file · response 1, 2

JSON
[
  {
    "type": "url_citation",
    "end_index": 301,
    "start_index": 196,
    "title": "Files | OpenAI API Reference",
    "url": "https://platform.openai.com/docs/api-reference/files?lang=ruby&utm_source=openai"
  }
]

Copied from Captured file (JSON), raw_response › output 3 › content 1 › annotations

Expected vs. Found
✗Expected: At least one returned result is a page that genuinely states the answer, 512 MB per file; any page that states it counts.Found: terra returned one result, also its only citation: a Files API reference link. That link lands on the Files index page, which gives no per-file size figure.
Supporting proof
Proof 1Screenshot
Where the cited link leads, 9 Oct 2026: the Files index page, no 512 MB figure
Open original ↗
Proof 2Captured file (JSON)
Fetch record, 28 Sep 2026: the cited URL refused our automated fetch (HTTP 403)
Cited values4 values from this file
citations 1 › url
https://platform.openai.com/docs/api-reference/files?lang=ruby&utm_source=openai
citations 1 › http_status
403
citations 1 › state
blocked
citations 1 › error
HTTP 403 Forbidden
Show the whole file (2 KB)
JSON
{
  "tc": "TC1",
  "run": "2026-09-28-r3-terra",
  "verified_on": "2026-09-11",
  "question": "What is the maximum size of an individual file I can upload through OpenAI's Files API?",
  "answer_text": "The maximum size for an individual file uploaded through OpenAI’s **Files API** is **512 MB**.  \n\nSome purposes impose tighter limits—for example, Batch API input files are limited to **200 MB**. ([platform.openai.com](https://platform.openai.com/docs/api-reference/files?lang=ruby&utm_source=openai))",
  "claim": "individual uploads to the OpenAI Files API may be up to 512 MB",
  "answer_key": "Individual files can be up to 512 MB, and each project can store up to 2.5 TB of files in total.",
  "n_citations": 1,
  "citations": [
    {
      "n": 1,
      "url": "https://platform.openai.com/docs/api-reference/files?lang=ruby&utm_source=openai",
      "title_from_tavily": "Files | OpenAI API Reference",
      "http_status": 403,
      "final_url": "https://platform.openai.com/docs/api-reference/files?lang=ruby&utm_source=openai",
      "redirected": false,
      "content_type": "text/html",
      "bytes": 10237,
      "page_title": "",
      "state": "blocked",
      "error": "HTTP 403 Forbidden",
      "primary_token_present": false,
      "secondary_token_present": false,
      "supports": "unjudged",
      "snippets": [],
      "snapshot": "_round2/evidence/citations/TC1/01-platform.openai.com.html"
    }
  ],
  "method": "Every citation URL opened with a browser user-agent, redirects followed, raw page snapshotted. Token presence is raw material only; the support judgement is made by reading the snippets, per V1's rule that it is a meaning judgement and never a word match."
}
Open the JSON file ↗
Proof 3Screenshot
Browser capture of the cited link, 14 Oct 2026: a Cloudflare verifying check, not the page
Open original ↗
Why this result

A returned page had to state the 512 MB limit, and any page that does counts. The only result, also the only citation, was a link whose landing page lists the Files endpoints with no file-size figure, and nothing else came back.

Also observed in this test case

The answer itself had 512 MB. terra's answer gave 512 MB per file and added that Batch API input files are limited to 200 MB; it is the source it returned, not the answer, that lacks the figure.

Sources without text. terra returned its one source as a link and title with no page text, so what the cited page says can only be checked by opening it.

One site-restricted search. Before answering, terra ran a single search of its own, limited to OpenAI's documentation site.

Tested by Aman · evidence dated 28 September 2026

Configuration and setup

How this tool was set up for the run and what the test needed in place. Each row is a fact from the run's records; a fact the records do not hold is left out, not guessed.

Software that produced the output
GPT-5.6 terra
Build or version
gpt-5.6-terra
Surface
API
Set up before the run
OpenAI's Upload file reference page (developers.openai.com/api/reference/resources/files/methods/create) gives the answer, 512 MB per file; it was checked by hand before the run and again on the run morning. The question went unchanged through the Responses API's web search tool at vendor defaults.
Tools
Web search (default settings: medium search context, text results)
Tested
28 September 2026 · Aman

How this scenario is graded

How we decide Pass, Fail and Not gradable. The same rules apply to every tool tested on this scenario.

How results are decided

Each test case gets one result per tool: Pass, Fail or Not gradable. A test case we haven't run yet shows Untested. There are no partial results.

The rules
  • Pass: the tool did everything the test expected, and nothing it said contradicts the correct answer.
  • Fail: at least one expected behaviour clearly didn't happen; the test case's result says which one and quotes the tool.
  • Not gradable: our evidence couldn't settle the outcome (for example a record we needed is missing). It is never counted as a fail, and the test case's result says what's missing.

Where this sits in the benchmark

This page is one cell of a larger study: one tool, one scenario. Only this benchmark's frame appears here.

LevelNameScope
BenchmarkWeb Search for AI →10 scenarios · 12 tools
CapabilityWeb Retrieval →
ScenarioA page on the web answers the question →
ToolGPT-5.6 terra →

History of this result

What has happened to this result since it was first published. Runs and grades are never overwritten: a retest or a re-grade publishes a new result and keeps the earlier one readable.

from the publication record
20 November 2026First publishedWeb Search for AI v1

Act on this result

Nothing filed here edits the run or the grade. A challenge opens a review, and a review can produce a new run or a re-grade — which becomes the current result and leaves this one in the history.

This matches what I see

You run the same kind of test against your own setup and get the same behaviour.

Agree →
This does not match

Yours behaves differently. Tell us what you got, with a screenshot if you have one.

Disagree →
Point out an issue

Something here is wrong — a reference value, a transcription, a grade.

Report an issue →
Request a retest

On a newer build, a larger dataset, or your own setup.

Request a retest →
We have fixed this

Tell us what changed and we schedule a rerun of the failing test case. The old result stays as history.

Vendor notice →
Filed against this evidenceNothing yet. Challenges, counter-evidence and fix notices appear here with their outcome, and stay on the page after they are resolved.
Cite this result
aidemos.com/benchmarks/web-search-for-ai/results/gpt-5-6-terra/a-page-on-the-web-answers-the-question · 0 pass · 1 fail · coverage 1/1 · graded 2026-11-19

The same record is available as structured data through the AI Demos MCP server, with the counts, the coverage and every per-test-case reason carried as fields.

Verify the proof files

These files support this result. Open a file to inspect the original evidence.

Captured file (JSON) ↗application/json · 7 KB
Proof 1 · Screenshot ↗image/png · 102 KB
Proof 2 · Captured file (JSON) ↗application/json · 2 KB
Proof 3 · Screenshot ↗image/png · 18 KB
File fingerprints (SHA-256)

A fingerprint identifies the exact file used for this result.

Captured file (JSON)75b46c64e0d9e71538f16798a9a6480b8e53981fb1c0ec199a5d98227aa1d8f2
Proof 1 · Screenshotbd13f41a4dfe6e679479d18fb4f0af26503096020edf98cae51231eb85f00370
Proof 2 · Captured file (JSON)15f344246053dc42dc98b04e236e237c04f8c5a42c405658d482efdae0b1a3a5
Proof 3 · Screenshot9a2268f55a6677125a0daed91946fabec0c5602ef665814ab584a09a8ac39a0d