///WebTaskBench
VERTICAL BENCHMARK · NEWS & MEDIA

News & Media

Major news publishers serving AI agents raw HTML at enormous token cost.

Leading news sites waste up to 78× more tokens than needed — TechCrunch, NYT, and others measured with Plasmate v0.5.0.

Avg Compression
39.9x
Peak Compression
77.5x
TechCrunch
Sites Measured
10

News & Media Rankings

Sorted by compression ratio · 10 sites

RANKSITEHTML TOKENSSOM TOKENSCOMPRESSION
1TechCrunchtechcrunch.comPeak compression — previously blocked by Cloudflare; fixed in Plasmate v0.5.0108K1K
77.5x
2The New York Timesnytimes.com293K5K
59.9x
3Washington Postwashingtonpost.com195K5K
38.2x
4BBCbbc.co.uk148K4K
37.9x
5Ars Technicaarstechnica.com155K4K
36.9x
6Financial Timesft.com138K4K
36.3x
7Wiredwired.com173K5K
33.9x
8Reutersreuters.com142K4K
33.0x
9The Vergetheverge.com168K5K
32.3x
10The Guardiantheguardian.com366K27K
13.4x

Cost Savings Estimate

Based on GPT-4o input pricing ($2.50/1M tokens) · News & Media average compression

5,000 pages/day
Est. Monthly Savings
$68285.81

What This Means

News sites are among the most frequently visited pages by AI agents performing research, summarization, and monitoring tasks. The average news publisher sends 41× more data than an agent actually needs — meaning every article lookup costs 41× more than necessary. For publishers serving millions of agent requests per day, switching to SOM could cut AI-related infrastructure costs by over 95%.

Ready to serve News & Media content efficiently?

Get started with SOM-ready pages today.

somready.com →