Claude Fable 5.2 Scores 73.3% to Gemini 4 Carbon's 71.9% on Google's Office-Task Benchmark
TLDR
Google's own benchmark report puts Claude Fable 5.2 at 73.3% against Gemini 4 Carbon's 71.9% across 200 visual office tasks, a gap the report treats as noise.
Optimixed Overview
Optimixed's original report on this story.
Google's own report on a 200-task visual office benchmark puts Claude Fable 5.2 at 73.3% and Gemini 4 Carbon at 71.9%, a difference the report treats as noise, as first reported by PPC Land. The piece also questions whether a vendor-built test can reliably guide AI agent buyers.
How we got here: earlier coverage
- Nov 10Google Makes Gemini 4 Carbon Available to Subscribers in Search AI Mode
Google released Gemini 4 Carbon, which is now available to subscribers in Google Search through AI Mode.
- Oct 23Google's Gemini 4 Argon overwhelms compute infrastructure with 25 trillion tokens consumed in 72 hours
Google's Gemini 4 Argon reportedly consumed 25 trillion tokens in its first 72 hours, straining compute infrastructure as demand runs 3x higher than Gemini 3.
- Oct 19Cloudflare claims Clairic model achieves 49% cost reduction in AI pipeline testing
Cloudflare announced a new Clairic AI model claiming 49% cost reduction in pipeline processing, though real-world applicability to SEO workflows remains unclear.
- Oct 13Google prices Gemini 4 Argon at $2 per million input tokens, output at $10M, restricts free access
Google released pricing for Gemini 4 Argon, a paid-only model with input costs of $2 per million tokens, output at $10M, and aggressive discounts for cached input.