Gemini 3.7 Flash ranks #57 of 87 models on the November 2026 ToneBench board through its best configuration, Gemini 3.7 Flash (high thinking). This setting, Gemini 3.7 Flash (medium thinking), is #118 of 151 configurations, with a Writing Score of 69.4/100 and a Writing Elo of 1014. One script costs $0.038 at API list price.
Last updated , from board data generated at 17:52 UTC.
Rank, scores, the nine metrics, the 10 reference scripts, cost and setup for one
configuration (a model at one reasoning setting). Key Results sums it up; the section links below it jump to the detail.
Model Rank ?
#57 of 87 modelsthrough Gemini 3.7 Flash (high thinking); this setting is #118 of 151 configurations
Writing Score ?
69.4 out of 100
Writing Elo ?
1014 95% range 967 to 1063
Cost per Script
$0.038 at API list price
Key Results
StandingGemini 3.7 Flash ranks #57 of 87 models through its best configuration, Gemini 3.7 Flash (high thinking). This setting is #118 of 151 configurations by Writing Elo (1014), in the bottom third of the board. Its 95% range (967 to 1063) overlaps 12 other configurations, ranked #110 to #122.
MetricsBest-ranked: Visual Cue Quality (74.1, 84th of 151 configurations, -0.6 against the board median). Lowest-ranked: Length Adherence (62.9, 123rd of 151 configurations, -21.8 against the board median).
ScriptsFrom 62.0 on Script 10 (engineering process walkthrough) to 74.7 on Script 8 (product release announcement): a 12.7-point spread across the 10 reference scripts.
ConsistencyIts Writing Score moves ± 3.6 between drafts of the same script, against ± 2.5 for the median configuration: more variable than most.
ValueGPT-6 Luna (medium) ranks far higher and costs about $0.0026 per script, against $0.038 per script for this one.
Louis's PickI'd stay with this one: Gemini 3.7 Flash (high thinking) is the strongest Gemini 3.7 Flash setting, but their Elo ranges overlap, and it costs about 5 cents more per script.
More Context Behind These Results
Writing Elo comes from comparing its average on each script with every other configuration's average on that same script; a gap smaller than the draft-to-draft variation counts as a draw (what the ranges can and cannot show).
Length Adherence ranks in the bottom half (123rd of 151 configurations), so plan on trimming or padding to reach the target length.
The script format moves its score a lot. If most of your scripts look like its best or weakest format, that result is a better guide than the average.
Two drafts of the same script can differ noticeably, so generating a few drafts and keeping the best pays off.
It outranks 3 configurations that cost at least 50% more.
Among proprietary models, Gemini 3.7 Flash is 42nd of 53. This setting is 95th of 108 proprietary configurations. The best open-weight configuration, GLM-5.3 (max) at configuration #22, outranks it.
Lab
Google 6th of 12 modelsthis setting 13th of 20 configurations
Weights
Proprietary 42nd of 53 modelsthis setting 95th of 108 configurations
Consistency
± 3.6 more variable than most
Average Tokens
21.3k input + output
Time per Script
0.9 min average, retries included
Highest-Ranked in This LabGemini 4 Carbon (high thinking) ranks highest of the Google configurations by Writing Elo, with a Writing Score +10.7 against this configuration, for about 55 cents more per script.
Each corner is one metric on a 0 to 100 scale; the further out, the higher the score. A corner outside the dashed line beats the board median. Hover or tap a point for its exact score, and a metric name for what it rewards.
Gemini 3.7 Flash (medium thinking)Board medianBoard best per metricMax possible (100)
Scores by Metric
Nine metrics, each scored 0 to 100 and weighted into the Writing Score; the weight sits under each name. Rank is out of all 151 ranked configurations.
Score colors:Strong: 82 and upMiddle: 65 to 81.9Weak: under 65
This modelBoard best (upper line)Board median (lower line)±: spread between drafts?: what the metric rewards
Tone & Voice Match ?19% weight · -13.1 vs median
board best 82.8 · Claude Opus 5.5 (max effort)60.9± 2.4 · 120thboard median 74.0
Writing Craft & Clarity ?13% weight · -7.3 vs median
board best 83.2 · Claude Opus 5.5 (xhigh)67.7± 1.3 · 118thboard median 75.0
Substance, Accuracy & Value ?15% weight · -8.9 vs median
board best 88.4 · Claude Opus 5.5 (xhigh)72.9± 2.5 · 115thboard median 81.8
Continuity & Emotion ?14% weight · -4.0 vs median
board best 85.7 · Claude Opus 5.5 (max effort)73.5± 1.5 · 104thboard median 77.5
YouTube Best Practices ?12% weight · -3.9 vs median
board best 84.5 · Claude Fable 5.2 (max effort)72.4± 2.3 · 102ndboard median 76.4
Hook Strength ?10% weight · -5.4 vs median
board best 84.0 · Claude Fable 5.2 (xhigh)74.2± 2.5 · 115thboard median 79.6
Length Adherence ?8% weight · -21.8 vs median
board best 99.0 · Claude Fable 5.2 (xhigh)62.9± 13.2 · 123rdboard median 84.8
Slop Score ?5% weight · -6.8 vs median
board best 89.0 · GPT-6 Sol (medium)74.5± 2.1 · 119thboard median 81.3
Visual Cue Quality ?4% weight · -0.6 vs median
board best 85.9 · Claude Opus 5.5 (max effort)74.1± 3.0 · 84thboard median 74.8
Scores by Reference Script
Every ranked configuration writes five drafts for each of the same 10 briefs, and each score is the mean of those drafts. The rank and board median compare it with every configuration on that script, which separates a weak result from a hard script. The result on the format closest to your own work is the most relevant.
Script 1
opinion / warning explainer
67.7
out of 100
122nd of 151 configurations on this script · board median 77.2
Script 2
news-analysis / skeptical explainer
72.0
out of 100
113th of 151 configurations on this script · board median 78.4
Script 3
personal roadmap / opinion
69.7
out of 100
117th of 151 configurations on this script · board median 77.2
Script 4
short explainer
68.8
out of 100
111th of 151 configurations on this script · board median 75.7
Script 5
news-analysis / opinion explainer
68.8
out of 100
117th of 151 configurations on this script · board median 78.0
Script 6
founder announcement / personal origin story
72.6
out of 100
121st of 151 configurations on this script · board median 79.2
Script 7
personal technical walkthrough / agentic workflow case study
70.3
out of 100
109th of 151 configurations on this script · board median 77.9
Script 8
product release announcement / personal observation
74.7
out of 100
109th of 151 configurations on this script · board median 79.0
Script 9
career guide / hiring analysis
67.7
out of 100
119th of 151 configurations on this script · board median 79.3
Script 10
engineering process walkthrough / presentation adaptation
62.0
out of 100
125th of 151 configurations on this script · board median 75.0
Consistency Between Drafts
Every score on this page is the mean of 5 drafts per script. The ± is the standard deviation between those drafts: the larger it is, the more two drafts of the same script can differ.
Compared with the typical model, it varies most on Length Adherence: ± 13.2 between drafts against ± 6.6, so two drafts of the same script can score quite differently there.
Consistency by Metric
Metric
This model
Board median
Verdict
Tone & Voice Match ?
60.9 ± 2.4
± 1.9
typical
Writing Craft & Clarity ?
67.7 ± 1.3
± 1.5
typical
Substance, Accuracy & Value ?
72.9 ± 2.5
± 2.1
typical
Continuity & Emotion ?
73.5 ± 1.5
± 1.5
typical
YouTube Best Practices ?
72.4 ± 2.3
± 2.4
typical
Hook Strength ?
74.2 ± 2.5
± 2.7
typical
Length Adherence ?
62.9 ± 13.2
± 6.6
more variable than most
Slop Score ?
74.5 ± 2.1
± 1.6
typical
Visual Cue Quality ?
74.1 ± 3.0
± 2.9
typical
Cost, Time and Tokens per Script
Logged on every run; none of these figures affects the Writing Score. Time is the average time to one accepted script, retries included. Cost is one script at the API list price.
Time per Script
0.9 minfaster than the board median of 1.5 min
Prompt tokens in
13.9kstyle guide + brief + research packet
Tokens Out
7.4kscript + any reasoning tokens
Cost per Script
$0.038measured from actual billed tokens; varies ± 0.020 between drafts
List Price Used
$0.75 / $3.75input / output, per million tokens
Scores by Judge
Four judges score the same 50 stored drafts blind against the same rubric: three LLM judges from three labs, plus Jev, a typed judge from TypeSafe that rates each metric on five described quality levels. No single judge's taste decides the result. The Writing Score of 69.4 combines their scores with the parts computed in code (Length Adherence and half of the Slop Score). The highest and lowest judge are 17.2 points apart on its Writing Score. That is close to the board median of 13.4. Each judge also has its own scale, so every card shows that judge's score against its average across the board. How the panel works.
Claude Opus 5.5 (medium)
Anthropic
62.7
-6.0 against its board average of 68.7
GLM 5.3 Flash
Z.ai
79.8
-3.0 against its board average of 82.9
GPT-6.1 Sol (medium)
OpenAI
65.4
-6.5 against its board average of 71.9
Jev 1.13
TypeSafe
69.9
-4.1 against its board average of 74.0
Evaluation Setup
How this exact configuration was run and judged: route, model ID, dates, sampling and prompt.
Exact model / route idgemini-3.7-flash (the request sets no level; the provider applies medium thinking).
Generations5 independent runs × 10 scripts = 50 drafts, stored after generation and never edited.
Run dates2026-10-11.
JudgesClaude Opus 5.5 at medium effort through the Claude Code subscription, GPT-6.1 Sol at medium reasoning through the Codex CLI, GLM 5.3 Flash through OpenRouter, and Jev 1.13 from TypeSafe. One call per judge per draft. All four score every draft on their own, blind to which model wrote it. The three LLM judges use the identical rubric and have to quote evidence before they score a metric. Jev answers the same rubric as one question per metric, choosing among five described quality levels, and writes no text. Seven metrics are the plain average of the four judges. Length is computed from the spoken word count, and Slop is half computed (lexical scans minus the spoken-dash penalty) and half judged. The Writing Elo is cross-checked with a Bradley-Terry fit of the same pairwise results (methodology).
Sampling and promptThe provider's default sampling, with no fine-tuning and no system-prompt changes. Every current-ranked model gets the same prompt, about 13,000 tokens (the count shifts a bit with each model's tokenizer): a style guide built from Louis's voice and storytelling guides, the target length, and the brief with its research packet. Generation is closed-book on every route: the prompt is the only source, with no web search, browsing, code execution, file access or other tool (generation).
What stays privateThe exact rubric and any non-public reference material. Each script is normally benchmarked before its video is published, which lowers the odds that a model has seen the reference, though it does not rule it out (why).
Frequently Asked Questions
How Good Is Gemini 3.7 Flash (medium thinking) at Writing?
Gemini 3.7 Flash ranks #57 of 87 models on ToneBench through its best configuration, Gemini 3.7 Flash (high thinking). This configuration is #118 of 151 configurations, in the bottom third of the board, with a Writing Elo of 1014 and a Writing Score of 69.4/100. The ranks come from full YouTube scripts written in the voice of Louis-François Bouchard and judged blind, with the channel's finished script as the reference. The benchmark measures long-form scripted writing, so the result does not carry over to every kind of text.
Is Gemini 3.7 Flash (medium thinking) the Best Google Model for Writing?
No. Gemini 3.7 Flash is 6th of 12 Google models. The top one is Gemini 4 Carbon, through Gemini 4 Carbon (high thinking), about 835 Elo ahead of this setting.
Is Gemini 3.7 Flash (medium thinking) Good Value for the Money?
It costs about $0.038 per script. GPT-6 Luna (medium) ranks higher and costs less.
What Are the Strengths and Weaknesses of Gemini 3.7 Flash (medium thinking)?
Its best-ranked metric is Visual Cue Quality (74.1, 84th of 151 configurations, -0.6 against the board median), and its lowest-ranked is Length Adherence (62.9, 123rd of 151, -21.8). Its best-ranked metric still sits in the bottom half. The Scores by Metric section shows the board median and the board's best for each one.
How Was Gemini 3.7 Flash (medium thinking) Evaluated?
It was run through the Google Gemini API, with the exact model/route ID gemini-3.7-flash, first run on 2026-10-11. It wrote each script 5 times with the provider's default sampling and no fine-tuning. The Evaluation Setup section and the methodology page have the details.
Does Gemini 3.7 Flash (medium thinking) Have Open Weights?
No. It is a proprietary model: its weights are not public. Among proprietary configurations it is 95th of 108 for writing.