Benmark · reviewed evidence

Creative tagging: benchmark methods and results · 16 October 2026

Models label image and video ads using a defined taxonomy. The published score averages tag-field correctness per output; critical-field checks distinguish a mostly correct record from one suitable for downstream analysis.

Reviewed 16 October 2026 · Tests and editorial direction: Ben Nancarrow · AI-assisted writing and checks.

Read the findings →

Download CSV Download JSON Permanent citation Latest reviewed evidence

Inputs, prompt conditions and scoring

What the model receives

The fixed media pack contains five images and five videos. Models return structured tags such as format, brand, product category, offer and CTA. Coverage varies across the published rows, so the results table retains each model's recorded output count.

What changes between prompts

Two prompt conditions are defined. Baseline receives allowed values and the JSON shape; the scaffold adds full definitions and supporting-evidence requirements for specified positive tags. Published model scores currently combine setups, with no complete paired comparison available on the site.

How the result is scored

Each output's score is its fraction of correctly resolved applicable fields, using exact checks and a semantic matcher where specified. The percentage shown here is the published mean 0–10 score multiplied by ten: mean correctness per output, not a pooled field-weighted accuracy or the percentage of entirely correct ads. The critical fields include format, brand, product category, offer and CTA; a high average can still contain a consequential error.

All 17 reviewed model results

Mean tag-field correctness per output (%)
Model	Result / setup	Scored outputs	Configuration / recorded test date	Cost / scored output
Claude Fable 5.1provisional	97.3%Combined setups	1	thinking_max2 September 2026	$0.384API-equivalent estimate
Meta Muse Wonderprovisional	95.9%Combined setups	20	max13 October 2026	$0.063Recorded spend
Gemini 3.5 Flashprovisional	95.1%Combined setups	120	reasoning_high, reasoning_medium5 July 2026	$0.004Recorded spend
Google Gemini 4 Argonprovisional	94.9%Combined setups	20	high13 October 2026	$0.046Recorded spend
Gemini 3.8 Flashprovisional	94.7%Combined setups	1	reasoning_high4 September 2026	$0.026Recorded spend
GPT-6 Astraprovisional	94.7%Combined setups	2	xhigh5 September 2026	$0.075Recorded spend
Claude Opus 5.5provisional	94.3%Combined setups	20	thinking_max23 September 2026	$0.339API-equivalent estimate
Gemini 3.1 Pro Previewprovisional	94.2%Combined setups	60	reasoning_high28 August 2026	$0.027Recorded spend
Gemini 3.7 Flashprovisional	94.2%Combined setups	60	reasoning_high20 August 2026	$0.004Recorded spend
Claude Opus 4.8provisional	93.7%Combined setups	180	thinking_high, thinking_medium, thinking_xhigh5 July 2026	$0.045Recorded spend
Claude Opus 5provisional	93.6%Combined setups	60	thinking_max20 August 2026	$0.061Recorded spend
GPT-6 Solprovisional	93.3%Combined setups	20	xhigh23 September 2026	$0.017Recorded spend
GPT-5.5provisional	92.5%Combined setups	60	reasoning_xhigh20 August 2026	$0.098Recorded spend
GPT-5.6 Solprovisional	92.4%Combined setups	60	xhigh20 August 2026	$0.029Recorded spend
Claude Fable 5provisional	92.1%Combined setups	20	thinking_max20 August 2026	$0.156API-equivalent estimate
Meta Muse Spark 1.3provisional	91.9%Combined setups	20	high8 September 2026	$0.019Recorded spend
Google: Gemma 4 26B A4Bprovisional	89.7%Combined setups	20	default13 June 2026	Not recorded

Recorded output counts can include repeats and both prompt conditions. Configurations and cost bases are retained; joint scores remain joint.

Complete paired prompt comparisons

No complete paired prompt comparison is available for this reviewed task.

Benmark (16 October 2026). Best LLM for creative tagging. https://benmark.ai/benchmarks/creative-tagging/releases/ed362c76c7f8706a6f19e3bb/

This record preserves the evidence and interpretation reviewed together. The live explorer can advance independently. Public results include provisional evidence; the methods above define what the scores measure.