Towards AITowards AIToneBench

Lab Comparison

OpenAI Models Compared

Last updated , from board data generated at 17:52 UTC.

All 32 ranked OpenAI configurations on ToneBench, side by side: Writing Elo, Writing Score, the nine metrics, and cost and time per script. A configuration is one model at one thinking level.

Highest Writing Elo
Elo 2301GPT-6 Sol (max)#5 of 87 models · $0.126 per script
Lowest Cost Within 200 Elo of the Top
$0.056 per scriptGPT-6 Sol (medium)Configuration #21 · Elo 2146
Lowest Cost
$0.0026 per scriptGPT-6 Luna (medium)Configuration #80 · Elo 1571

A configuration 200 Elo below the top is expected to score about 24% in its script-by-script comparisons with it, a draw counting as half. The Lowest Cost Within 200 Elo of the Top card names the cheapest configuration that stays that close.

Recommended Configurations

Ranks in these notes are configuration ranks: each setting's place among all 151 configurations. The Highest Writing Elo card above gives the model rank.

Picks and Notes From Louis

The best OpenAI script comes from GPT-6 Sol (max), at $0.126 and 2.5 minutes per script. That's cheap for a model at #13, so most people can stop there. For volume, I'd run the same model at medium: GPT-6 Sol (medium) is #21, 155 Elo back, for $0.056 and 0.8 minutes per script. It's also the cheapest configuration within 200 Elo of the top.

The higher version number doesn't help you here. GPT-6.1 Sol (max) lands at #18, behind GPT-6 Sol (max), and it costs more and takes 10.2 minutes per script. The new GPT-6.1 Terra, OpenAI's cheaper tier below Sol, doesn't write better than GPT-5.6 Terra either, but below max it's far cheaper and faster, and its effort settings have similar Writing Elo ratings. So I'd run it at low for cheap first drafts in bulk, #50 for $0.027 in about 0.5 minutes, and step up to GPT-6 Sol (medium), my volume pick, when the script matters: it costs 2.1 times as much and ranks far higher. The one I'd skip is GPT-6 Astra (max). It costs 9.1 times what GPT-6 Sol (max) does and ranks lower, so for writing I don't see where it fits.

On a tight budget, GPT-6 Luna (max) gets you #70 for $0.0046 a script, which is a lot of writer for that price. I'd leave the GPT-5.5 settings alone, because they miss the target length badly, and GPT-4o is near the very bottom of the board now. The open gpt-oss rows are down around #138, so I wouldn't pick them for scripts either.

OpenAI is a hard family to pick from, just because there's so much of it: 32 configurations across 13 models, from GPT-4o mini up to GPT-6.1 Sol, most of the recent ones at several effort settings. Keeping Sol, Terra, Luna and Astra straight is a job in itself. You won't get the top script on the board here, since the best OpenAI row is GPT-6 Sol (max) at #13, but the top of the family is packed tight and mostly cheap. The bottom is older and smaller models, which is how you get 2333 Elo between first and last. So the OpenAI name alone tells you very little. One disclosure: GPT-6.1 Sol is also one of the four judges, and the other three score every draft too.

All 32 OpenAI Configurations, Grouped by Model

Every row wrote the same 10 scripts for the same four blind judges, so these Writing Elo ratings sit on the leaderboard's scale. The rank column is the configuration rank, each setting's place among all 151 configurations on the board. Rows are grouped by model: a highlighted row is a model's best configuration, and the rows under it are the same model at other thinking levels. The ± after each Elo is half the width of its 95% range, and scores run from 0 to 100. Cost is one full script at the API list price of the route used. Each name opens its scorecard, and the head-to-head comparisons cover models from different labs.

Score colors:82 and up65 to 81.9under 65

With the metric scores shown, the table scrolls sideways.

Configuration RankConfigurationWriting EloWriting ScoreCost / ScriptTime / ScriptTone & Voice MatchWriting Craft & ClaritySubstance, Accuracy & ValueHook StrengthYouTube Best PracticesContinuity & EmotionSlop ScoreLength AdherenceVisual Cue Quality
#13GPT-6 Sol (max)2301 ±4083.5$0.1262.5 min80.280.587.983.280.782.388.592.981.5
#21GPT-6 Sol (medium)2146 ±4282.5$0.0560.8 min80.180.686.282.779.381.989.088.178.2
#18GPT-6.1 Sol (max)2153 ±3682.4$0.23910.2 min78.178.187.382.878.679.288.298.083.0
#23GPT-6.1 Sol (xhigh)2136 ±4182.4$0.1997.8 min77.778.187.383.278.779.288.098.082.1
#31GPT-6.1 Sol (low)2009 ±3481.4$0.0521.5 min77.778.686.283.579.180.588.283.881.5
#33GPT-6.1 Sol (high)1975 ±3981.1$0.0943.3 min76.977.886.682.178.279.587.887.881.6
#37GPT-6.1 Sol (medium)1946 ±3481.0$0.0541.6 min77.278.285.881.878.980.288.085.180.5
#24GPT-5.6 Sol (ultra)2134 ±5282.3$0.3674.1 min77.778.188.083.080.181.585.692.081.6
#25GPT-5.6 Sol (low)2132 ±3582.3$0.1130.9 min78.679.387.183.779.981.785.489.779.9
#27GPT-6 Astra (max)2113 ±4482.3$1.159.2 min77.878.487.582.176.679.988.698.383.4
#41GPT-6 Astra (medium)1918 ±3680.7$0.2621.5 min77.078.486.082.576.680.088.583.280.7
#32GPT-5.6 Terra (ultra)1993 ±4281.4$0.2758.0 min76.976.886.481.579.380.284.793.780.3
#40GPT-5.6 Terra (medium)1922 ±3980.7$0.0621.4 min78.678.084.681.777.879.783.487.177.2
#46GPT-6.1 Terra (max)1884 ±3580.3$0.0791.8 min74.078.385.880.676.377.588.194.280.2
#49GPT-6.1 Terra (high)1880 ±3880.3$0.0270.5 min75.378.785.481.576.479.087.988.677.0
#50GPT-6.1 Terra (low)1878 ±4580.2$0.0270.5 min75.278.785.181.476.378.988.088.076.7
#53GPT-6.1 Terra (xhigh)1860 ±4080.1$0.0280.5 min75.278.685.680.876.579.087.886.877.8
#55GPT-6.1 Terra (medium)1847 ±4579.9$0.0270.5 min75.778.785.081.676.278.988.083.877.0
#65GPT-5.6 Luna (ultra)1731 ±4178.8$0.0263.1 min71.974.784.880.676.677.283.991.878.4
#85GPT-5.6 Luna (medium)1533 ±4176.7$0.00650.5 min73.275.583.682.270.977.183.070.976.1
#70GPT-6 Luna (max)1662 ±3878.2$0.00461.3 min71.377.084.080.873.377.286.086.877.4
#75GPT-6 Luna (high)1604 ±5077.4$0.00280.6 min71.076.582.680.869.277.085.789.174.4
#80GPT-6 Luna (medium)1571 ±5277.0$0.00260.5 min71.376.882.380.568.477.185.284.874.9
#87GPT-5.5 (xhigh)1530 ±4276.8$0.1840.9 min78.375.284.580.776.977.780.651.378.1
#90GPT-5.5 (high)1516 ±4776.4$0.1830.9 min78.175.084.481.377.277.580.646.978.5
#93GPT-5.5 (medium)1460 ±4675.8$0.1761.0 min78.975.484.380.977.677.780.935.978.7
#97GPT-5.5 (none)1421 ±5475.2$0.1740.8 min78.375.083.980.377.577.280.233.178.8
#109GPT-5.2 (no reasoning)1109 ±4571.4$0.0811.0 min75.070.082.377.874.873.777.620.177.8
#138gpt-oss 120B (high) · open weights541 ±4461.3$0.00731.6 min52.061.356.168.464.262.165.676.959.4
#140gpt-oss 120B (medium) · open weights507 ±5260.1$0.00360.5 min52.659.557.969.461.960.864.069.452.6
#148GPT-4o (no thinking)4 ±3747.3$0.0440.2 min40.450.750.959.845.850.053.129.648.2
#149GPT-4o mini (no thinking)-32 ±3447.0$0.00280.2 min39.348.753.762.340.248.152.344.428.9

Effort Ladders for OpenAI Models

Each of these models ran at more than one thinking level. Each page shows how the score, cost and time change from one level to the next.

GPT-6 Sol Against Other OpenAI Models

Each comparison lists every thinking level of both models, with Writing Elo, cost and time side by side. The line under each pair gives the Writing Elo gap between the two best configurations. A pair outside the written match-ups opens in the comparison tool.

← All comparisons