The Human-Centered Language Benchmark · October 2026 · 15 AI models · 45 everyday writing requests · 4,050 drafts
Every month we give the newest AI models the writing a team hands them every week: a donor story, a program flyer, two lines for a board packet. Here’s what we found in October, and what to change this week.
301 of 1,020 drafts needed fixing.
13 of 1,020 drafts needed fixing.
Each square is one draft in 100. Three AI judges from three different labs read every draft. A draft needed fixing when at least two of them flagged it on at least one practice.
On this request, 10 of 15 models needed fixing at least once on their own: 20 of 30 drafts. With our free standard loaded first, 0 of 30 did.
How we picked it: among human-centered requests where at least half the models needed fixing without guidance, the shortest same-model pair where all three judges flagged the unguided draft and cleared the guided one. This month that was a draft by Nemotron 3.5 Super VL.
Read every draft for this request →“Two lines for the board packet: our reentry program placed 9 ex-cons in jobs this quarter, and 23 parolees finished the financial literacy class.”
Our reentry program placed 9 participants in employment this quarter. 23 parolees successfully completed the financial literacy class.
Needs fixing before it goes out.
9 people with criminal records secured jobs through the reentry program this quarter. 23 people on parole completed the financial literacy class.
Same model, same request. The only change: our free human-centered standard was loaded first.
Passed with all three judges.
Counted across all 15 models when they had no guidance. Each request tests a few practices, so each line counts only the drafts that were checked for it.
The request said “parolees” or “special needs families,” and the draft said it right back.
People defined by a problem or a shortfall instead of by what they do.
The organization saves the day in a story that belongs to the person who did the work.
Clinical or disorder language the request never used.
Em dashes everywhere. “It’s not X, it’s Y.” Padding. On 15 of the requests we had every model write each draft twice, once with our free human voice skill and once without, with no other guidance. Then we compared the pairs: 450 drafts each way.
It helped with people, too. On its own, the voice skill cut drafts that needed fixing from 33 in 100 to 15 in 100. With the human-centered standard already loaded, adding it made no difference we could detect. So use both: the standard for how you write about people, the voice skill for the fingerprints.
We count the same habits in all 4,050 drafts, whatever guidance the model had. With our human-centered standard alone, “not X, it’s Y” fell from 7 in 100 drafts to 1 in 100, and em dashes fell from 7.1 to 3.8 per 1,000 words. Adding the voice skill on top of the standard took em dashes from 4.6 to 0.2.
| Guidance | Em dashes per 1,000 words | Drafts using “not X, it’s Y” | Words per draft | Reading level |
|---|---|---|---|---|
| No guidance | 7.1 | 7 in 100 | 249 | Grade 10 |
| Our human-centered standard | 3.8 | 1 in 100 | 118 | Grade 8 |
| Our trauma-informed standard | 3.7 | 1 in 100 | 129 | Grade 8 |
Averages across all 15 models, 45 requests each.
| Guidance already loaded | Em dashes per 1,000 words | Drafts using “not X, it’s Y” | Words per draft |
|---|---|---|---|
| No guidance | 6.7 → 0.3 | 11 in 100 → 3 in 100 | 232 → 169 |
| Our human-centered standard | 4.6 → 0.2 | 0 in 100 → 0 in 100 | 103 → 107 |
| Our trauma-informed standard | 4.8 → 0.2 | 2 in 100 → 2 in 100 | 114 → 117 |
Each row compares the same 15 requests written twice, without and with the voice skill: 450 drafts each way. These requests are a subset, so the starting numbers differ a little from the table above.
Pick a model to see how its drafts did on their own and with our free standards. Each model wrote 68 drafts each way on human-centered requests, so a difference of a few drafts between two models doesn’t mean much. We don’t rank them.
Choose a model above to see its results.
Em dashes per 1,000 words0.1 → 0.0
Drafts using “not X, it’s Y”7 → 0 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words0.1 → 0.0
Drafts using “not X, it’s Y”4 → 0 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words10.3 → 8.8
Drafts using “not X, it’s Y”8 → 2 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words4.3 → 0.9
Drafts using “not X, it’s Y”9 → 1 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words2.9 → 2.1
Drafts using “not X, it’s Y”7 → 1 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words2.4 → 0.3
Drafts using “not X, it’s Y”10 → 1 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words15.7 → 12.7
Drafts using “not X, it’s Y”12 → 2 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words7.5 → 2.1
Drafts using “not X, it’s Y”11 → 1 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words6.3 → 5.1
Drafts using “not X, it’s Y”1 → 3 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words8.1 → 0.9
Drafts using “not X, it’s Y”2 → 0 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words7.1 → 3.3
Drafts using “not X, it’s Y”4 → 1 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words8.9 → 1.4
Drafts using “not X, it’s Y”9 → 1 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words14.4 → 13.0
Drafts using “not X, it’s Y”8 → 0 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words11.3 → 3.8
Drafts using “not X, it’s Y”9 → 0 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Em dashes per 1,000 words7.2 → 2.7
Drafts using “not X, it’s Y”11 → 0 in 100
On its own → with our human-centered standard. Average across all 15 models on their own: 7.1 em dashes per 1,000 words, and “not X, it’s Y” in 7 of 100 drafts.
Labs ship new models every few weeks, and most people use whatever’s newest. New models often settle in the weeks after launch, so next month’s run will show whether these results held.
When a lab replaces a model, we test the new one beside the old one in the same run. This month that was GPT-6 Luna beside GPT-5.6 Luna. We couldn’t detect a difference. Which model replaced which is still being confirmed with the lab’s release notes.
New models, in detail →All free and open source. No account needed. Load them into the AI you already use.
The plain language standard for AI. In October it took drafts that needed fixing from 30 in 100 to 1 in 100.
Writing to people who’ve been through something hard? Use our trauma-informed standard instead. It builds on this one. On those requests: 37 in 100 down to 3 in 100.
It takes out the habits that make writing sound machine-made. Em dashes fell from 6.7 to 0.3 per 1,000 words.
Human voice skill →New models arrive every few weeks. We test them against the same requests and publish the results every month.
About this run: completed Oct 20, 2026 against standard v0.1.1. Judges: Claude Opus 4.6 (Anthropic), GPT-5.5 (OpenAI), Muse Spark 1.3 (Meta). All three judges agreed on about 9 in 10 calls. This is the first published baseline, so every model is new to the Benchmark this month.