Storia in 2 fonti

Inception Labs' Mercury 2 AI Beats Google's DiffusionGemma at Its Own Game - Decrypt

Both models trade word-by-word generation for parallel denoising. Only one of them does it without losing intelligence in the trade.

Raccontata da

decrypt.co

cryptobriefing.com

Confronto fonti

2 prospettive sulla stessa storia

AI · summaries

decrypt.coStai leggendo2 mesi fa

Inception Labs' Mercury 2 AI Beats Google's DiffusionGemma at Its Own Game - Decrypt

Mercury 2 generates 1,000 tokens/sec and scores 90% on AIME 2026 vs. DiffusionGemma's 69.1% via parallel diffusion. Augment Code cut sub-agent latency 82% and cost 90% with Mercury 2—proving parallel generation scales agentic production systems.

originale

cryptobriefing.com2 mesi fa

Inception Labs' Mercury 2 outperforms Google's DiffusionGemma in the race to replace autoregressive AI

Mercury 2 (Inception Labs, Feb 2026) reaches 1,009 tokens/sec at $0.25/$0.75 per million tokens, outperforming Google's DiffusionGemma while preserving complex reasoning in parallel generation. Diffusion inference redefines GPU workload economics and disrupts current budget-tier pricing.

Inception Labs' Mercury 2 AI Beats Google's DiffusionGemma at Its Own Game - Decrypt

Confronto fonti

Inception Labs' Mercury 2 AI Beats Google's DiffusionGemma at Its Own Game - Decrypt

Inception Labs' Mercury 2 outperforms Google's DiffusionGemma in the race to replace autoregressive AI

Timeline cronologica

Inception Labs' Mercury 2 AI Beats Google's DiffusionGemma at Its Own Game - Decrypt

Inception Labs' Mercury 2 outperforms Google's DiffusionGemma in the race to replace autoregressive AI

Inception Labs' Mercury 2 AI outperforms Google's DiffusionGemma: DecryptMedia