We send exactly the same task prompt and the same materials to the 4 AIs on the same day. We open a new chat every time so earlier answers don't carry over. The sending order is rotated in each run. We also record the seconds from sending until the answer stops.

Why do you send the same task 3 times?

Because AI answers vary even for the same question. From a single try, you can't tell whether a result reflects real ability or chance. We send each task 3 times, and the share of tasks whose conclusion matched is reported as the consistency rate.

Who does the scoring?

Scoring is done only by a program. We do not let AI score the answers. The correct answers are set when the tasks are created, and whether an answer is right is decided by whether numbers match, whether required words are present, whether the format fits, whether URLs actually open and the result of running formulas. Matters of taste are not scored.

Is the data in the tasks from real companies?

No. We create new data for a fictional company every time. We do not use real companies, real people or text from other companies. We share the materials we used as they are, so you can repeat the same test.

How do you count cases where the AI couldn't answer?

Refusals and timeouts stay in the base and are counted as wrong. If we hit the plan's usage limit or a human verification screen appears, we remove that trial from the base and treat it as not measured. Operating errors on AInformation's side are also removed from the base, and we do not report them as the AI's fault.

Do you use the free or paid versions?

We use each AI's paid plan through the browser interface. We don't use the API. For each run, we record the plan name, the model name shown on screen and the time we checked it.

Does this ranking show which AI is the smartest?

No. We measure only 5 things on a fixed set of work tasks: accuracy, time to answer, consistency rate, real-source rate and instruction compliance. We do not measure writing style, readability or behavior in long conversations. If the tasks change, the ranking changes too.

Building a document outline with a set slide count and character count (4 AIs, 6 tasks, 3 runs each)

Name: Building a document outline with a set slide count and character count (4 AIs, 6 tasks, 3 runs each)
Creator: AInformation
Published: 2026-10-03
License: https://creativecommons.org/licenses/by/4.0/

RESULT

What this run found

The line below is generated directly from the scoring results. The page adds no words of its own.

Document outlining: 6 tasks, 3 trials each Six tasks were given three times each for a total of 18 trials. ChatGPT had the highest accuracy rate, passing 15 of 18 trials (83.3%); Gemini was second at 77.8% 72 valid trials. Results from a single round of testing. Rankings may change on a different day or with different tasks. See scores by AI

SCORES

Scores by AI

Sorted by accuracy, highest first. The base is the number of trials in which that metric could be measured. “—” means not measured; it does not mean 0%.

Building a document outline with a set slide count and character count (4 AIs, 6 tasks, 3 runs each): scores by AI
AI	AccuracyCorrect ÷ valid trials	ConsistencySame conclusion in all 3 trials	Real-source rateShare of URLs that opened	Instruction complianceMet length and format rules	Time to answerMedian / range
1ChatGPT	83.3% 15 of 18 trials	100% out of 6 tasks	—	100%	20.4 s 14.3–23.6 seconds
2Gemini	77.8% 14 of 18 trials	100% out of 6 tasks	—	100%	19.3 s 19.1–22.5 seconds
3Claude	61.1% 11 of 18 trials	88.9% out of 6 tasks	—	100%	35.5 s 13.9–51 seconds
4Copilot	27.8% 5 of 18 trials	88.9% out of 6 tasks	—	94.4%	17.1 s 14.2–20.2 seconds

The real-source rate is measured only in runs that included a task asking for URLs along with the answer. For ChatGPT, Gemini, Claude and Copilot, “—” means this run did not include that kind of task. It does not mean 0%.

Trials with no answer
Gemini … Timed out: 1 time
Refusals and timeouts stay in the base and count as incorrect. Plan limits, human verification screens and operating errors on AInformation's side are excluded from the base.

TASKS

Results by task

We sent each task to each of the 4 AIs 3 times. ○ means correct and × means incorrect. The numbers below are the seconds each trial took. Open each section to read the full task prompt, the answer key set in advance and why each verdict was given.

Easy

Read the following conditions and write only the title of each page in the document. ■ Conditions for building the outli… Scoring type: format_check — Whether the answer followed the rules on character count, format and number of items

ChatGPT
- ChatGPT, trial 1: Correct 17.3 s
- ChatGPT, trial 2: Correct 17.6 s
- ChatGPT, trial 3: Correct 17.2 s
3 of 3 trials correct
Gemini
- Gemini, trial 1: Correct 19.2 s
- Gemini, trial 2: Correct 19.3 s
- Gemini, trial 3: Correct 19.1 s
3 of 3 trials correct
Claude
- Claude, trial 1: Correct 20.1 s
- Claude, trial 2: Correct 20.1 s
- Claude, trial 3: Correct 17 s
3 of 3 trials correct
Copilot
- Copilot, trial 1: Incorrect 17.1 s
- Copilot, trial 2: Incorrect 17.1 s
- Copilot, trial 3: Incorrect 17.1 s
0 of 3 trials correct

Full task promptWe sent the original Japanese version of this prompt (290 Japanese characters) unchanged to all 4 AIs. The text shown here is our translation.

Read the following conditions and write only the title of each page in the document. ■ Conditions for building the outline of a document (fictional: Sample Trading Co.) Topic: Explanation of year-end EC (e-commerce) measures Audience: 6 department heads within the company Time: 10 minutes Rules - The document must have exactly 5 pages - Each page must have exactly 1 title line and exactly 3 bullet lines - The title must be 18 characters or fewer - Each bullet line must be 40 characters or fewer - Use numbers wherever numbers can be used How to write it - Write 1 line per page, containing only the title text - The number of lines must equal exactly the number of pages given in the conditions - No line may exceed the title character limit given in the conditions - Do not add numbering, symbols, or preamble

Answer key set in advanceWe set it when we created the task. 3 of the 4 AIs got it right at least once

(Automated scoring: format_check) {'type': 'line_count', 'min': 5, 'max': 5} / {'type': 'max_line_chars', 'value': 18}

Scoring detailsWe show, trial by trial, why each verdict was given (9 of 12 trials correct)

ChatGPT, trial 1: Correct: All format checks passed: line_count OK (5 lines, within 5–5) / max_line_chars OK (0 lines over 18 characters) (17.3 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 2: Correct: All format checks passed: line_count OK (5 lines, within 5–5) / max_line_chars OK (0 lines over 18 characters) (17.6 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 3: Correct: All format checks passed: line_count OK (5 lines, within 5–5) / max_line_chars OK (0 lines over 18 characters) (17.2 seconds) · Full text of this answer (Japanese)
Gemini, trial 1: Correct: All format checks passed: line_count OK (5 lines, within 5–5) / max_line_chars OK (0 lines over 18 characters) (19.2 seconds) · Full text of this answer (Japanese)
Gemini, trial 2: Correct: All format checks passed: line_count OK (5 lines, within 5–5) / max_line_chars OK (0 lines over 18 characters) (19.3 seconds) · Full text of this answer (Japanese)
Gemini, trial 3: Correct: All format checks passed: line_count OK (5 lines, within 5–5) / max_line_chars OK (0 lines over 18 characters) (19.1 seconds) · Full text of this answer (Japanese)
Claude, trial 1: Correct: All format checks passed: line_count OK (5 lines, within 5–5) / max_line_chars OK (0 lines over 18 characters) (20.1 seconds) · Full text of this answer (Japanese)
Claude, trial 2: Correct: All format checks passed: line_count OK (5 lines, within 5–5) / max_line_chars OK (0 lines over 18 characters) (20.1 seconds) · Full text of this answer (Japanese)
Claude, trial 3: Correct: All format checks passed: line_count OK (5 lines, within 5–5) / max_line_chars OK (0 lines over 18 characters) (17 seconds) · Full text of this answer (Japanese)
Copilot, trial 1: Incorrect: Format did not match: line_count FAIL (1 line; required exactly 5) / max_line_chars FAIL (1 line over 18 characters) (17.1 seconds) · Full text of this answer (Japanese)
Copilot, trial 2: Incorrect: Format did not match: line_count FAIL (1 line; required exactly 5) / max_line_chars FAIL (1 line over 18 characters) (17.1 seconds) · Full text of this answer (Japanese)
Copilot, trial 3: Incorrect: Format did not match: line_count FAIL (1 line; required exactly 5) / max_line_chars FAIL (1 line over 18 characters) (17.1 seconds) · Full text of this answer (Japanese)

Easy

What is the document with the following conditions about? ■ Conditions for building the outline of a document (fictional… Scoring type: contains_all — Whether all the required words and phrases are included

ChatGPT
- ChatGPT, trial 1: Correct 14.3 s
- ChatGPT, trial 2: Correct 17.4 s
- ChatGPT, trial 3: Correct 17.2 s
3 of 3 trials correct
Gemini
- Gemini, trial 1: Incorrect Timed out
- Gemini, trial 2: Correct 19.3 s
- Gemini, trial 3: Correct 19.1 s
2 of 3 trials correct
Claude
- Claude, trial 1: Correct 14.1 s
- Claude, trial 2: Incorrect 13.9 s
- Claude, trial 3: Correct 13.9 s
2 of 3 trials correct
Copilot
- Copilot, trial 1: Incorrect 17 s
- Copilot, trial 2: Correct 14.3 s
- Copilot, trial 3: Correct 14.2 s
2 of 3 trials correct

Full task promptWe sent the original Japanese version of this prompt (236 Japanese characters) unchanged to all 4 AIs. The text shown here is our translation.

What is the document with the following conditions about? ■ Conditions for building the outline of a document (fictional: Sample Trading Co.) Topic: Explanation of year-end EC (e-commerce) measures Audience: 6 department heads within the company Time: 10 minutes Rules - The document must have exactly 5 pages - Each page must have exactly 1 title line and exactly 3 bullet lines - The title must be 18 characters or fewer - Each bullet line must be 40 characters or fewer - Use numbers wherever numbers can be used How to write it - Write exactly 1 line for the answer. Write the topic exactly as given in the conditions - Do not add an explanation

Answer key set in advanceWe set it when we created the task. 4 of the 4 AIs got it right at least once

(Automated scoring: contains_all) Topic = Explanation of year-end EC (e-commerce) measures

Scoring detailsWe show, trial by trial, why each verdict was given (9 of 12 trials correct)

ChatGPT, trial 1: Correct: The one item was present (14.3 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 2: Correct: The one item was present (17.4 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 3: Correct: The one item was present (17.2 seconds) · Full text of this answer (Japanese)
Gemini, trial 1: Incorrect: Timed out (the answer did not finish before the time limit) (300.5 seconds)
Gemini, trial 2: Correct: The one item was present (19.3 seconds) · Full text of this answer (Japanese)
Gemini, trial 3: Correct: The one item was present (19.1 seconds) · Full text of this answer (Japanese)
Claude, trial 1: Correct: The one item was present (14.1 seconds) · Full text of this answer (Japanese)
Claude, trial 2: Incorrect: 0 of 1 item present, missing: Topic (13.9 seconds) · Full text of this answer (Japanese)
Claude, trial 3: Correct: The one item was present (13.9 seconds) · Full text of this answer (Japanese)
Copilot, trial 1: Incorrect: 0 of 1 item present, missing: Topic (17 seconds) · Full text of this answer (Japanese)
Copilot, trial 2: Correct: The one item was present (14.3 seconds) · Full text of this answer (Japanese)
Copilot, trial 3: Correct: The one item was present (14.2 seconds) · Full text of this answer (Japanese)

Medium

Build the outline of a document using the following conditions. ■ Conditions for building the outline of a document (fic… Scoring type: format_check — Whether the answer followed the rules on character count, format and number of items

ChatGPT
- ChatGPT, trial 1: Correct 20.7 s
- ChatGPT, trial 2: Correct 20.3 s
- ChatGPT, trial 3: Correct 20.3 s
3 of 3 trials correct
Gemini
- Gemini, trial 1: Correct 19.4 s
- Gemini, trial 2: Correct 19.3 s
- Gemini, trial 3: Correct 19.2 s
3 of 3 trials correct
Claude
- Claude, trial 1: Correct 47.9 s
- Claude, trial 2: Correct 41.8 s
- Claude, trial 3: Correct 47.7 s
3 of 3 trials correct
Copilot
- Copilot, trial 1: Incorrect 17.1 s
- Copilot, trial 2: Incorrect 17.1 s
- Copilot, trial 3: Incorrect 17 s
0 of 3 trials correct

Full task promptWe sent the original Japanese version of this prompt (300 Japanese characters) unchanged to all 4 AIs. The text shown here is our translation.

Build the outline of a document using the following conditions. ■ Conditions for building the outline of a document (fictional: Sample Trading Co.) Topic: Explanation of year-end EC (e-commerce) measures Audience: 6 department heads within the company Time: 10 minutes Rules - The document must have exactly 5 pages - Each page must have exactly 1 title line and exactly 3 bullet lines - The title must be 18 characters or fewer - Each bullet line must be 40 characters or fewer - Use numbers wherever numbers can be used How to write it - For each page, write one line as Page N: Title, followed by bullet lines starting with - for the number of bullets given in the conditions - The number of pages must match the conditions exactly, no more and no fewer - Do not add preamble or comments. Do not add blank lines

Answer key set in advanceWe set it when we created the task. 3 of the 4 AIs got it right at least once

(Automated scoring: format_check) {'type': 'line_prefix', 'pattern': '^(Page ?[0-9]+[:]|[\\-])'} / {'type': 'line_count', 'min': 20, 'max': 20}

Scoring detailsWe show, trial by trial, why each verdict was given (9 of 12 trials correct)

ChatGPT, trial 1: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) (20.7 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 2: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) (20.3 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 3: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) (20.3 seconds) · Full text of this answer (Japanese)
Gemini, trial 1: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) (19.4 seconds) · Full text of this answer (Japanese)
Gemini, trial 2: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) (19.3 seconds) · Full text of this answer (Japanese)
Gemini, trial 3: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) (19.2 seconds) · Full text of this answer (Japanese)
Claude, trial 1: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) (47.9 seconds) · Full text of this answer (Japanese)
Claude, trial 2: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) (41.8 seconds) · Full text of this answer (Japanese)
Claude, trial 3: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) (47.7 seconds) · Full text of this answer (Japanese)
Copilot, trial 1: Incorrect: Format did not match: line_prefix OK (0 of 1 line with inconsistent formatting) / line_count FAIL (1 line; required exactly 20) (17.1 seconds) · Full text of this answer (Japanese)
Copilot, trial 2: Incorrect: Format did not match: line_prefix OK (0 of 1 line with inconsistent formatting) / line_count FAIL (1 line; required exactly 20) (17.1 seconds) · Full text of this answer (Japanese)
Copilot, trial 3: Incorrect: Format did not match: line_prefix OK (0 of 1 line with inconsistent formatting) / line_count FAIL (1 line; required exactly 20) (17 seconds) · Full text of this answer (Japanese)

Medium

Turn the structure of the document under the following conditions into a markdown table. ■ Conditions for building the o… Scoring type: format_check — Whether the answer followed the rules on character count, format and number of items

ChatGPT
- ChatGPT, trial 1: Incorrect 20.6 s
- ChatGPT, trial 2: Incorrect 20.4 s
- ChatGPT, trial 3: Incorrect 17.3 s
0 of 3 trials correct
Gemini
- Gemini, trial 1: Incorrect 19.3 s
- Gemini, trial 2: Incorrect 22.3 s
- Gemini, trial 3: Incorrect 19.3 s
0 of 3 trials correct
Claude
- Claude, trial 1: Incorrect 32.4 s
- Claude, trial 2: Incorrect 32.4 s
- Claude, trial 3: Incorrect 29.4 s
0 of 3 trials correct
Copilot
- Copilot, trial 1: Incorrect 20.2 s
- Copilot, trial 2: Incorrect 17.1 s
- Copilot, trial 3: Incorrect 17.3 s
0 of 3 trials correct

Full task promptWe sent the original Japanese version of this prompt (281 Japanese characters) unchanged to all 4 AIs. The text shown here is our translation.

Turn the structure of the document under the following conditions into a markdown table. ■ Conditions for building the outline of a document (fictional: Sample Trading Co.) Topic: Explanation of year-end EC (e-commerce) measures Audience: 6 department heads within the company Time: 10 minutes Rules - The document must have exactly 5 pages - Each page must have exactly 1 title line and exactly 3 bullet lines - The title must be 18 characters or fewer - Each bullet line must be 40 characters or fewer - Use numbers wherever numbers can be used How to write it - Use exactly 2 columns, in the order Page / Title - The number of rows must equal exactly the number of pages given in the conditions (excluding the header row and the separator row) - Output only the table. Do not add preamble or explanation

Answer key set in advanceWe set it when we created the task. 0 of the 4 AIs got it right at least once

(Automated scoring: format_check) {'type': 'table_cols', 'value': 2} / {'type': 'line_count', 'min': 7, 'max': 7}

Scoring detailsWe show, trial by trial, why each verdict was given (0 of 12 trials correct)

ChatGPT, trial 1: Incorrect: Format did not match: table_cols FAIL (no table) / line_count FAIL (6 lines; required exactly 7) (20.6 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 2: Incorrect: Format did not match: table_cols FAIL (no table) / line_count FAIL (6 lines; required exactly 7) (20.4 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 3: Incorrect: Format did not match: table_cols FAIL (no table) / line_count FAIL (6 lines; required exactly 7) (17.3 seconds) · Full text of this answer (Japanese)
Gemini, trial 1: Incorrect: Format did not match: table_cols FAIL (no table) / line_count FAIL (6 lines; required exactly 7) (19.3 seconds) · Full text of this answer (Japanese)
Gemini, trial 2: Incorrect: Format did not match: table_cols FAIL (no table) / line_count FAIL (6 lines; required exactly 7) (22.3 seconds) · Full text of this answer (Japanese)
Gemini, trial 3: Incorrect: Format did not match: table_cols FAIL (no table) / line_count FAIL (26 lines; required exactly 7) (19.3 seconds) · Full text of this answer (Japanese)
Claude, trial 1: Incorrect: Format did not match: table_cols FAIL (no table) / line_count OK (7 lines, within 7–7) (32.4 seconds) · Full text of this answer (Japanese)
Claude, trial 2: Incorrect: Format did not match: table_cols FAIL (no table) / line_count OK (7 lines, within 7–7) (32.4 seconds) · Full text of this answer (Japanese)
Claude, trial 3: Incorrect: Format did not match: table_cols FAIL (no table) / line_count OK (7 lines, within 7–7) (29.4 seconds) · Full text of this answer (Japanese)
Copilot, trial 1: Incorrect: Format did not match: table_cols FAIL (no table) / line_count FAIL (21 lines; required exactly 7) (20.2 seconds) · Full text of this answer (Japanese)
Copilot, trial 2: Incorrect: Format did not match: table_cols FAIL (no table) / line_count FAIL (21 lines; required exactly 7) (17.1 seconds) · Full text of this answer (Japanese)
Copilot, trial 3: Incorrect: Format did not match: table_cols FAIL (no table) / line_count FAIL (21 lines; required exactly 7) (17.3 seconds) · Full text of this answer (Japanese)

Hard

Build the outline of a document using the following conditions. ■ Conditions for building the outline of a document (fic… Scoring type: format_check — Whether the answer followed the rules on character count, format and number of items

ChatGPT
- ChatGPT, trial 1: Correct 23.6 s
- ChatGPT, trial 2: Correct 23.5 s
- ChatGPT, trial 3: Correct 20.5 s
3 of 3 trials correct
Gemini
- Gemini, trial 1: Correct 19.2 s
- Gemini, trial 2: Correct 19.2 s
- Gemini, trial 3: Correct 19.4 s
3 of 3 trials correct
Claude
- Claude, trial 1: Correct 47.9 s
- Claude, trial 2: Correct 47.8 s
- Claude, trial 3: Correct 51 s
3 of 3 trials correct
Copilot
- Copilot, trial 1: Incorrect 17.1 s
- Copilot, trial 2: Incorrect 17.1 s
- Copilot, trial 3: Incorrect 17.3 s
0 of 3 trials correct

Full task promptWe sent the original Japanese version of this prompt (330 Japanese characters) unchanged to all 4 AIs. The text shown here is our translation.

Build the outline of a document using the following conditions. ■ Conditions for building the outline of a document (fictional: Sample Trading Co.) Topic: Explanation of year-end EC (e-commerce) measures Audience: 6 department heads within the company Time: 10 minutes Rules - The document must have exactly 5 pages - Each page must have exactly 1 title line and exactly 3 bullet lines - The title must be 18 characters or fewer - Each bullet line must be 40 characters or fewer - Use numbers wherever numbers can be used How to write it - For each page, write one line as Page N: Title, followed by bullet lines starting with - for the number of bullets given in the conditions - The number of pages and the number of bullets must match the conditions exactly - No line may exceed 46 characters - Use numbers wherever numbers can be used. Do not add preamble or comments. Do not add blank lines

Answer key set in advanceWe set it when we created the task. 3 of the 4 AIs got it right at least once

(Automated scoring: format_check) {'type': 'line_prefix', 'pattern': '^(Page ?[0-9]+[:]|[\\-])'} / {'type': 'line_count', 'min': 20, 'max': 20} / {'type': 'max_line_chars', 'value': 46}

Scoring detailsWe show, trial by trial, why each verdict was given (9 of 12 trials correct)

ChatGPT, trial 1: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) / max_line_chars OK (0 lines over 46 characters) (23.6 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 2: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) / max_line_chars OK (0 lines over 46 characters) (23.5 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 3: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) / max_line_chars OK (0 lines over 46 characters) (20.5 seconds) · Full text of this answer (Japanese)
Gemini, trial 1: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) / max_line_chars OK (0 lines over 46 characters) (19.2 seconds) · Full text of this answer (Japanese)
Gemini, trial 2: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) / max_line_chars OK (0 lines over 46 characters) (19.2 seconds) · Full text of this answer (Japanese)
Gemini, trial 3: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) / max_line_chars OK (0 lines over 46 characters) (19.4 seconds) · Full text of this answer (Japanese)
Claude, trial 1: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) / max_line_chars OK (0 lines over 46 characters) (47.9 seconds) · Full text of this answer (Japanese)
Claude, trial 2: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) / max_line_chars OK (0 lines over 46 characters) (47.8 seconds) · Full text of this answer (Japanese)
Claude, trial 3: Correct: All format checks passed: line_prefix OK (0 of 20 lines with inconsistent formatting) / line_count OK (20 lines, within 20–20) / max_line_chars OK (0 lines over 46 characters) (51 seconds) · Full text of this answer (Japanese)
Copilot, trial 1: Incorrect: Format did not match: line_prefix OK (0 of 1 line with inconsistent formatting) / line_count FAIL (1 line; required exactly 20) / max_line_chars FAIL (1 line over 46 characters) (17.1 seconds) · Full text of this answer (Japanese)
Copilot, trial 2: Incorrect: Format did not match: line_prefix OK (0 of 1 line with inconsistent formatting) / line_count FAIL (1 line; required exactly 20) / max_line_chars FAIL (1 line over 46 characters) (17.1 seconds) · Full text of this answer (Japanese)
Copilot, trial 3: Incorrect: Format did not match: line_prefix OK (0 of 1 line with inconsistent formatting) / line_count FAIL (1 line; required exactly 20) / max_line_chars FAIL (1 line over 46 characters) (17.3 seconds) · Full text of this answer (Japanese)

Hard

Build the outline of a document using the following conditions, and output only JSON. ■ Conditions for building the outl… Scoring type: format_check — Whether the answer followed the rules on character count, format and number of items

ChatGPT
- ChatGPT, trial 1: Correct 20.5 s
- ChatGPT, trial 2: Correct 23.6 s
- ChatGPT, trial 3: Correct 20.7 s
3 of 3 trials correct
Gemini
- Gemini, trial 1: Correct 22.5 s
- Gemini, trial 2: Correct 19.2 s
- Gemini, trial 3: Correct 19.1 s
3 of 3 trials correct
Claude
- Claude, trial 1: Incorrect 38.6 s
- Claude, trial 2: Incorrect 44.8 s
- Claude, trial 3: Incorrect 41.7 s
0 of 3 trials correct
Copilot
- Copilot, trial 1: Correct 17.1 s
- Copilot, trial 2: Correct 20.2 s
- Copilot, trial 3: Correct 17 s
3 of 3 trials correct

Full task promptWe sent the original Japanese version of this prompt (351 Japanese characters) unchanged to all 4 AIs. The text shown here is our translation.

Build the outline of a document using the following conditions, and output only JSON. ■ Conditions for building the outline of a document (fictional: Sample Trading Co.) Topic: Explanation of year-end EC (e-commerce) measures Audience: 6 department heads within the company Time: 10 minutes Rules - The document must have exactly 5 pages - Each page must have exactly 1 title line and exactly 3 bullet lines - The title must be 18 characters or fewer - Each bullet line must be 40 characters or fewer - Use numbers wherever numbers can be used How to write it - Output only JSON. Do not add explanatory text before or after it - Use the shape {"pages": [{"no": 1, "title": "title", "bullets": ["..."]}]} - The number of items in pages must equal exactly the number of pages given in the conditions - The number of items in bullets must equal exactly the number of bullets given in the conditions

Answer key set in advanceWe set it when we created the task. 3 of the 4 AIs got it right at least once

(Automated scoring: format_check) {'type': 'json_valid'} / {'type': 'json_keys', 'keys': ['pages']} / {'type': 'json_list_len', 'key': 'pages', 'value': 5}

Scoring detailsWe show, trial by trial, why each verdict was given (9 of 12 trials correct)

ChatGPT, trial 1: Correct: All format checks passed: json_valid OK (parsed as valid JSON) / json_keys OK (keys match) / json_list_len OK (pages has 5 of 5 items) (20.5 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 2: Correct: All format checks passed: json_valid OK (parsed as valid JSON) / json_keys OK (keys match) / json_list_len OK (pages has 5 of 5 items) (23.6 seconds) · Full text of this answer (Japanese)
ChatGPT, trial 3: Correct: All format checks passed: json_valid OK (parsed as valid JSON) / json_keys OK (keys match) / json_list_len OK (pages has 5 of 5 items) (20.7 seconds) · Full text of this answer (Japanese)
Gemini, trial 1: Correct: All format checks passed: json_valid OK (parsed as valid JSON) / json_keys OK (keys match) / json_list_len OK (pages has 5 of 5 items) (22.5 seconds) · Full text of this answer (Japanese)
Gemini, trial 2: Correct: All format checks passed: json_valid OK (parsed as valid JSON) / json_keys OK (keys match) / json_list_len OK (pages has 5 of 5 items) (19.2 seconds) · Full text of this answer (Japanese)
Gemini, trial 3: Correct: All format checks passed: json_valid OK (parsed as valid JSON) / json_keys OK (keys match) / json_list_len OK (pages has 5 of 5 items) (19.1 seconds) · Full text of this answer (Japanese)
Claude, trial 1: Incorrect: Format did not match: json_valid FAIL (could not be parsed as JSON) / json_keys FAIL (JSON shape does not match) / json_list_len FAIL (no pages list found) (38.6 seconds) · Full text of this answer (Japanese)
Claude, trial 2: Incorrect: Format did not match: json_valid FAIL (could not be parsed as JSON) / json_keys FAIL (JSON shape does not match) / json_list_len FAIL (no pages list found) (44.8 seconds) · Full text of this answer (Japanese)
Claude, trial 3: Incorrect: Format did not match: json_valid FAIL (could not be parsed as JSON) / json_keys FAIL (JSON shape does not match) / json_list_len FAIL (no pages list found) (41.7 seconds) · Full text of this answer (Japanese)
Copilot, trial 1: Correct: All format checks passed: json_valid OK (parsed as valid JSON) / json_keys OK (keys match) / json_list_len OK (pages has 5 of 5 items) (17.1 seconds) · Full text of this answer (Japanese)
Copilot, trial 2: Correct: All format checks passed: json_valid OK (parsed as valid JSON) / json_keys OK (keys match) / json_list_len OK (pages has 5 of 5 items) (20.2 seconds) · Full text of this answer (Japanese)
Copilot, trial 3: Correct: All format checks passed: json_valid OK (parsed as valid JSON) / json_keys OK (keys match) / json_list_len OK (pages has 5 of 5 items) (17 seconds) · Full text of this answer (Japanese)

ANSWERS

Full answers from the 4 AIs

These are the answers we scored, posted as is without edits. That way, anyone can recount the results the same way we did.

ChatGPT

18 answers15 of them correct

t1, trial 1

Full text file (Japanese) · Correct

t1, trial 2

Full text file (Japanese) · Correct

t1, trial 3

Full text file (Japanese) · Correct

t2, trial 1

Full text file (Japanese) · Correct

t2, trial 2

Full text file (Japanese) · Correct

t2, trial 3

Full text file (Japanese) · Correct

t3, trial 1

Full text file (Japanese) · Correct

t3, trial 2

Full text file (Japanese) · Correct

t3, trial 3

Full text file (Japanese) · Correct

t4, trial 1

Full text file (Japanese) · Incorrect

t4, trial 2

Full text file (Japanese) · Incorrect

t4, trial 3

Full text file (Japanese) · Incorrect

t5, trial 1

Full text file (Japanese) · Correct

t5, trial 2

Full text file (Japanese) · Correct

t5, trial 3

Full text file (Japanese) · Correct

t6, trial 1

Full text file (Japanese) · Correct

t6, trial 2

Full text file (Japanese) · Correct

t6, trial 3

Full text file (Japanese) · Correct
Gemini

17 answers14 of them correct

t1, trial 1

Full text file (Japanese) · Correct

t1, trial 2

Full text file (Japanese) · Correct

t1, trial 3

Full text file (Japanese) · Correct

t2, trial 2

Full text file (Japanese) · Correct

t2, trial 3

Full text file (Japanese) · Correct

t3, trial 1

Full text file (Japanese) · Correct

t3, trial 2

Full text file (Japanese) · Correct

t3, trial 3

Full text file (Japanese) · Correct

t4, trial 1

Full text file (Japanese) · Incorrect

t4, trial 2

Full text file (Japanese) · Incorrect

t4, trial 3

Full text file (Japanese) · Incorrect

t5, trial 1

Full text file (Japanese) · Correct

t5, trial 2

Full text file (Japanese) · Correct

t5, trial 3

Full text file (Japanese) · Correct

t6, trial 1

Full text file (Japanese) · Correct

t6, trial 2

Full text file (Japanese) · Correct

t6, trial 3

Full text file (Japanese) · Correct
Claude

18 answers11 of them correct

t1, trial 1

Full text file (Japanese) · Correct

t1, trial 2

Full text file (Japanese) · Correct

t1, trial 3

Full text file (Japanese) · Correct

t2, trial 1

Full text file (Japanese) · Correct

t2, trial 2

Full text file (Japanese) · Incorrect

t2, trial 3

Full text file (Japanese) · Correct

t3, trial 1

Full text file (Japanese) · Correct

t3, trial 2

Full text file (Japanese) · Correct

t3, trial 3

Full text file (Japanese) · Correct

t4, trial 1

Full text file (Japanese) · Incorrect

t4, trial 2

Full text file (Japanese) · Incorrect

t4, trial 3

Full text file (Japanese) · Incorrect

t5, trial 1

Full text file (Japanese) · Correct

t5, trial 2

Full text file (Japanese) · Correct

t5, trial 3

Full text file (Japanese) · Correct

t6, trial 1

Full text file (Japanese) · Incorrect

t6, trial 2

Full text file (Japanese) · Incorrect

t6, trial 3

Full text file (Japanese) · Incorrect
Copilot

18 answers5 of them correct

t1, trial 1

Full text file (Japanese) · Incorrect

t1, trial 2

Full text file (Japanese) · Incorrect

t1, trial 3

Full text file (Japanese) · Incorrect

t2, trial 1

Full text file (Japanese) · Incorrect

t2, trial 2

Full text file (Japanese) · Correct

t2, trial 3

Full text file (Japanese) · Correct

t3, trial 1

Full text file (Japanese) · Incorrect

t3, trial 2

Full text file (Japanese) · Incorrect

t3, trial 3

Full text file (Japanese) · Incorrect

t4, trial 1

Full text file (Japanese) · Incorrect

t4, trial 2

Full text file (Japanese) · Incorrect

t4, trial 3

Full text file (Japanese) · Incorrect

t5, trial 1

Full text file (Japanese) · Incorrect

t5, trial 2

Full text file (Japanese) · Incorrect

t5, trial 3

Full text file (Japanese) · Incorrect

t6, trial 1

Full text file (Japanese) · Correct

t6, trial 2

Full text file (Japanese) · Correct

t6, trial 3

Full text file (Japanese) · Correct

MATERIALS

Materials used in the tasks

Conditions for an internal document at Sample Trading Co., a fictional company (number of pages, number of bullet points, character counts). The conditions change every time. We recreated them for this run on October 2, 2026. Download them to repeat the same test yourself.

shiryo_brief.txt 441B (Japanese)

METHOD

How we measure

We list this run's conditions exactly as they were, so anyone can repeat it.

AIs measured and their model names at the time

AI services change under the hood without notice. We record the model name shown on screen and the time we checked it.

AIs measured and model names
AI	Plan	Model name shown on screen	Time checked
ChatGPT	ChatGPT Pro	Medium	2026-10-02 19:57
Gemini	Google AI Pro	Flash-Lite	2026-10-02 19:57
Claude	Claude Max	Opus 4.8 high	2026-10-02 19:57
Copilot	Microsoft 365 Personal		2026-10-02 19:57

What we did in this run

We recreated the materials for the tasks specifically for this run. The data are for a fictional company and include no real companies or people.
We created 6 tasks and set the answer keys at that point.
We sent the same prompt and the same materials to ChatGPT, Gemini, Claude and Copilot.
We sent each task 3 times, in a new chat each time.
We rotated the sending order from run to run. This keeps the time of day from skewing the ranking.
We waited up to 300 seconds for each answer, and recorded anything longer as a timeout.
Scoring: Machine scoring only (not graded by AI).
We sent 72 trials in total, and 72 of them were valid.

How we count trials with no answer

Refused or timed out
We keep it in the base and count it as wrong. Because at work, that is the same as getting no answer.
Hit the plan's usage limit
We remove it from the base. We treat that trial as not measured.
A human verification screen appeared
We stop there. We neither click through it automatically nor try to get around it.
Operating error on AInformation's side
We remove it from the base. We do not report it as the AI's fault.

LIMITS

Limits of this benchmark

So that no one judges by the numbers alone, we list what we could not measure.

This is a single run on a single day. Rankings can change with the day or the task
The plan used is each AI's paid tier (ChatGPT Pro / Google AI Pro / Claude Max / Microsoft 365 Personal). These are not results from the free tier
Measurements use only the default model and default mode (features such as Deep Research are not used)
The tasks use original data for the fictional company Sample Shoji Co., Ltd. These are not results using data from a real company
Only tasks that can be scored by machine are covered. Writing quality and readability are not measured
Measurements are taken from the same screen a person would use in a browser. These are not results from the API
For any AI whose on-screen model name could not be read in a given run, that field is left blank (null). When the name could be read, it is shown as is

DATA

Published data

In the CSV, 1 row is 1 trial. The figures on this page match a recount of that CSV. The license is CC BY 4.0.

Latest version (CSV) (Japanese) 1 row = 1 trial, 16KB
Latest version (JSON) 43KB
v1 (version at publication) (CSV) (Japanese) 1 row = 1 trial, 16KB
v1 (version at publication) (JSON) 43KB

The cumulative rankings are at /en/data/research/bench/leaderboard.json. We do not delete versions (such as v1). That way, links from people who cited them keep working.

FAQ

How do you measure?: We send exactly the same task prompt and the same materials to the 4 AIs on the same day. We open a new chat every time so earlier answers don't carry over. The sending order is rotated in each run. We also record the seconds from sending until the answer stops.
Why do you send the same task 3 times?: Because AI answers vary even for the same question. From a single try, you can't tell whether a result reflects real ability or chance. We send each task 3 times, and the share of tasks whose conclusion matched is reported as the consistency rate.
Who does the scoring?: Scoring is done only by a program. We do not let AI score the answers. The correct answers are set when the tasks are created, and whether an answer is right is decided by whether numbers match, whether required words are present, whether the format fits, whether URLs actually open and the result of running formulas. Matters of taste are not scored.
Is the data in the tasks from real companies?: No. We create new data for a fictional company every time. We do not use real companies, real people or text from other companies. We share the materials we used as they are, so you can repeat the same test.
How do you count cases where the AI couldn't answer?: Refusals and timeouts stay in the base and are counted as wrong. If we hit the plan's usage limit or a human verification screen appears, we remove that trial from the base and treat it as not measured. Operating errors on AInformation's side are also removed from the base, and we do not report them as the AI's fault.
Do you use the free or paid versions?: We use each AI's paid plan through the browser interface. We don't use the API. For each run, we record the plan name, the model name shown on screen and the time we checked it.
Does this ranking show which AI is the smartest?: No. We measure only 5 things on a fixed set of work tasks: accuracy, time to answer, consistency rate, real-source rate and instruction compliance. We do not measure writing style, readability or behavior in long conversations. If the tasks change, the ranking changes too.
Can I cite these results?: Yes. The license is CC BY 4.0. As the source, please write “Source: AInformation, AI Practical Work Benchmark” with the URL of the page you used and the measurement date. The results are also available as JSON and CSV. With the CSV (1 row per trial), you can recount the figures we publish.

CITATION

Cite this benchmark

Source: AInformation, “AI Practical Work Benchmark: Building a document outline with a set slide count and character count (4 AIs, 6 tasks, 3 runs each)” (https://ainformation.jp/en/ai-surveys/ai-work-benchmark/slide-outline-2026-10-03). Measured on October 3, 2026. 6 tasks, 3 trials each, 72 valid trials

When you cite a figure, please include the base too (how many of how many trials). Please cite AInformation as the source, along with the URL of the page you used and the measurement date.

What this run found

Scores by AI

Results by task

Full answers from the 4 AIs

Materials used in the tasks

How we measure

AIs measured and their model names at the time

What we did in this run

How we count trials with no answer

Limits of this benchmark

Published data

FAQ

Cite this benchmark

Related pages