OPEN LINK ↗
// 1h agoMODEL RELEASE
Inception Mercury 3 Launches At 1,068 Tokens/Sec
Inception launched Mercury 3, a diffusion LLM claiming 1,068 tokens per second, a 1M-token context window, and 4–8× the speed of GPT-6 Luna and Claude Haiku 5.5. The launch also spotlights Mercury Decide, a separate decision model positioned against Jev.
// ANALYSIS
The real story is Inception betting that parallel denoising can make latency—and agent economics—a first-class model advantage. Mercury Decide suggests the company is building a portfolio around fast machine decisions, not just fast prose.
- –Diffusion generation could reduce latency across browser, search, voice, and support agents.
- –Mercury 3 reportedly scales beyond Mercury 1 while preserving 1,000+ tokens/sec.
- –The evidence is still mostly vendor-reported, without a full set of reproducible benchmark scores.
- –Mercury Decide’s JevBench and LegalBench positioning targets high-volume classification and routing.
- –OpenAI API compatibility makes latency-sensitive A/B testing relatively easy.
// TAGS
mercury-3mercury-decidellminferenceagentapilong-context
DISCOVERED
1h ago
2026-10-19
PUBLISHED
1h ago
2026-10-19
RELEVANCE
9/ 10
AUTHOR
EverymansAI
