Mercury 3 diffusion LLM tested: matches Haiku 5.5 quality at nearly 2x the speed
daniel_mac8 · x · 2026-10-20
Diffusion language models are back: Inception AI's Mercury 3 claims to match Claude Haiku 5.5 and GPT-6 Luna on average benchmark scores at 1,000+ tokens/s.
danielmac8 tested the same prompt on medium reasoning: Haiku 5.5 (via claude.ai) took 8.0s, Mercury 3 (via OpenRouter) just 4.3s — nearly 2x faster. The diffusion approach's speed advantage is becoming tangible.
More from Models
- Anthropic Safeguards blog: Claude self-assessment flags unsafe chats, 78% recall at 1% FPR — javirandor · 2026-10-20
- Unverified: Gemini 4 Argon reportedly catching up with Opus 5.5 — vedantmisra · 2026-10-20
- Researcher: OpenAI appears roughly two months ahead of Anthropic on capability trend — peterwildeford · 2026-10-20
- OpenCode ships DeepSeek V4.1 Flash 1017 update, zero changes needed for users — intellectronica · 2026-10-20
- Space Bunny revealed: MiniMax M3.1 Flash topped OpenCode for 12 straight days — MiniMax_AI · 2026-10-20
- MiniMax M3.1 Flash lands on Ollama with US hosting, zero data retention, 1M context — MiniMax_AI · 2026-10-20