Towards AITowards AIToneBench

GPT-5.6 Terra (low)

OpenAI · proprietary · writing benchmark

#32 of 130 Elo 2012 Overall 85.9 Proprietary
Rank
#32 of 130
Writing Elo
2012 ±45
Overall
85.9 / 100
Cost / task
$0.075 per script
Family
OpenAI 13th of 36
Type
Closed 30th of 83
Consistency
± 1.9 steadier than most
Avg tokens
15.0k in+out
Latency
46s per call
Family check:GPT-5.6 Sol (ultra) is OpenAI's best writer here, +2.5 overall vs this config for $0.062 more per script.

The short version

GPT-5.6 Terra (low) sits at #32 of 130 on ToneBench, in the middle of the pack, with a writing Elo of 2012 and an overall score of 85.9 out of 100. We measured it by having it write all 9 of our real YouTube scripts, five times each, then scoring every draft blind against our own finished versions. Here is how it shook out.

Within OpenAI it ranks 13th of 36. GPT-5.6 Sol (ultra) is the family's top writer here, about 304 Elo ahead, so this config trades some quality for whatever it saves you in cost or speed.

Among proprietary models, it comes in 30th of 83. That ranking is against the stronger half of the board: closed models still set the pace on voice fidelity here.

One thing that stands out: it is unusually consistent. Across its 45 drafts the overall score only moves ± 1.9 points run to run, versus a board median of ± 3.8. What you get on the first try is close to what you get every try.

What it does best is Anti-Slop: it ranks 18th on that dimension at 92.7, well above the board average. Its softer spot is Length Discipline (69.9, 69th), which is the thing to watch if that metric matters most for your use.

It was uneven across the 9 scripts. Its best run was the career guide / hiring analysis (88.5) and its weakest was the product release announcement / personal observation (83.8), a spread of about 5 points. Worth knowing if your writing skews toward one of those formats.

On cost, it runs about $0.075/task. Grok 4.5 scores higher for less money, at $0.038/task, so on pure value this config is not the frontier. Still, it beats 10 models that cost noticeably more.

If you're tuning reasoning effort, we also tested GPT-5.6 Terra at other settings. The strongest of those on the board is GPT-5.6 Terra (xhigh) (Elo 2110), so it is worth checking whether more or less thinking moves the needle before you lock in this one.

Skill profile

The shape is the story: the further a corner reaches, the stronger that dimension. Biggest edge over the board: Visual Cues (+14.1); smallest edge: Length Discipline (+2.7).

50Tone?Craft?Substance?Flow?YouTube?Hook?Length?Anti-Slop?Cues?
GPT-5.6 Terra (low)Board averageBoard best per metricMax possible (100)

Per-metric scores

Nine writing dimensions, each scored 0–100 and blended by the editorial weight shown. Rank is against all 130 current-ranked models. The thick bar is this model; the thin lines above and below are the current field's best model (darker) and average (lighter) on the same scale. The small ± number is its run-to-run variation.

Tone & Voice ?19% weight · +9.5 vs avg
best · Claude Opus 5 (max effort) · 91.287.8± 1.4 · 22ndboard avg · 78.3
Writing Craft ?13% weight · +8.4 vs avg
best · Claude Opus 5 (max effort) · 91.187.3± 1.6 · 25thboard avg · 78.9
Substance & Value ?15% weight · +9.7 vs avg
best · Claude Opus 5 (max effort) · 90.688.4± 1.9 · 26thboard avg · 78.7
Flow & Emotion ?14% weight · +10.3 vs avg
best · Claude Opus 5 (max effort) · 90.484.7± 1.8 · 26thboard avg · 74.4
YouTube Structure ?12% weight · +11.9 vs avg
best · Claude Opus 5 (max effort) · 89.685.2± 2.4 · 27thboard avg · 73.3
Hook ?10% weight · +6.9 vs avg
best · Claude Opus 5 (max effort) · 92.087.2± 1.9 · 28thboard avg · 80.2
Length Discipline ?8% weight · +2.7 vs avg
best · GPT-5.6 Sol (ultra) · 94.369.9± 7.2 · 69thboard avg · 67.2
Anti-Slop ?5% weight · +9.9 vs avg
best · Claude Fable 5 (max effort + 4.8 fallback) · 93.792.7± 1.4 · 18thboard avg · 82.8
Visual Cues ?4% weight · +14.1 vs avg
best · GPT-5.6 Sol (xhigh) · 91.488.7± 2.1 · 20thboard avg · 74.5

Per-article scores

The same 9 real scripts every current-ranked model writes, scored individually. Different formats stress different skills.

Article 1
opinion / warning explainer
86.6
out of 100
Article 2
news-analysis / skeptical explainer
86.5
out of 100
Article 3
personal roadmap / opinion
85.9
out of 100
Article 4
short explainer
86.4
out of 100
Article 5
news-analysis / opinion explainer
84.7
out of 100
Article 6
founder announcement / personal origin story
84.8
out of 100
Article 7
personal technical walkthrough / agentic workflow case study
85.5
out of 100
Article 8
product release announcement / personal observation
83.8
out of 100
Article 9
career guide / hiring analysis
88.5
out of 100

Consistency: run-to-run variability

Every cell on this page is the mean of 5 independent runs per script; the ± numbers are the run-to-run standard deviation. Overall, GPT-5.6 Terra (low) varies by ± 1.9 points between runs versus a board median of ± 3.8, so it is noticeably steadier than the typical model here.

Its steadiest is Anti-Slop (± 1.4 vs ± 2.6 board median), which you can rely on run after run.

MetricThis modelBoard medianVerdict
Tone & Voice ?87.8 ± 1.4± 2.2typical
Writing Craft ?87.3 ± 1.6± 2.0typical
Substance & Value ?88.4 ± 1.9± 2.6typical
Flow & Emotion ?84.7 ± 1.8± 2.6typical
YouTube Structure ?85.2 ± 2.4± 3.6typical
Hook ?87.2 ± 1.9± 3.0typical
Length Discipline ?69.9 ± 7.2± 8.6typical
Anti-Slop ?92.7 ± 1.4± 2.6steadier than most
Visual Cues ?88.7 ± 2.1± 4.7steadier than most

Measured, not modeled

Numbers we log on every run and rarely talk about. None of these affect the writing scores; cost and latency are informational.

Latency per script
46sfaster than the board median of 70s
Prompt tokens in
12.1kstyle guide + brief + research packet
Tokens out
3.0kscript + any reasoning tokens
Cost per script
$0.075 ± 0.028measured from actual billed tokens
List price used
$2.5 / $15 per M tokinput / output

What each judge scored it

The published overall of 85.9 is the consensus of three family-disjoint judges scoring the same 45 stored drafts blind with the identical rubric. The highest and lowest judge differ by 8.0 points on its overall. A typical amount of judge disagreement for this board. How the panel works: methodology.

Claude Opus 5
Anthropic
82.4
GPT-5.6 Sol (medium)
OpenAI
90.5
DeepSeek V4 Flash
DeepSeek
84.8

How we ran GPT-5.6 Terra (low)

Frequently asked questions

How good is GPT-5.6 Terra (low) at writing?

On ToneBench it ranks #32 of 130 with a writing Elo of 2012 and an overall score of 85.9/100. That score comes from writing our 9 real YouTube scripts five times each and scoring every draft blind against our own finished versions across nine writing dimensions.

Is GPT-5.6 Terra (low) the best OpenAI model for writing?

Not quite. Within OpenAI it ranks 13th of 36; GPT-5.6 Sol (ultra) is the family's best writer here.

Is GPT-5.6 Terra (low) good value for the money?

It costs about $0.075/task. Grok 4.5 scores higher for less, so it is not the value pick.

What are GPT-5.6 Terra (low)'s strengths and weaknesses?

Its strongest dimension is Anti-Slop (18th on the board, 92.7). Its weakest is Length Discipline (69th on the board, 69.9). The full nine-metric breakdown is on this page.

Does GPT-5.6 Terra (low) write some formats better than others?

Yes. Its best of our 9 scripts was the career guide / hiring analysis (88.5) and its weakest was the product release announcement / personal observation (83.8).

How consistent is GPT-5.6 Terra (low) between runs?

We run every script 5 times. GPT-5.6 Terra (low)'s overall score varies by about ±1.9 points between runs, versus a board median of ±3.8. That makes it one of the steadier models we test. The full per-metric variability table is on this page.

How was GPT-5.6 Terra (low) evaluated?

Via Codex CLI (OpenAI subscription) using the exact model/route id gpt-5.6-terra at reasoning_effort low, run on 2026-07-29. 5 runs per script, provider-default sampling, no fine-tuning; every draft scored blind with a fixed rubric by a three-family judge panel. Full details in the 'How we ran it' section and the methodology.

Is GPT-5.6 Terra (low) open source?

No, it is a proprietary (closed-weights) model. Among proprietary models it ranks 30th of 83 for writing.

← Back to the full ToneBench leaderboard