news

Inception releases Mercury 3, a diffusion language model it says runs up to 8x faster than GPT-6 Luna and Claude Haiku 5.5

Inception released Mercury 3, a diffusion-based LLM with a 1M-token context. In the company's own benchmarks it beat GPT-6 Luna and Claude Haiku 5.5 on 8 of 12 tests at 4–8x their speed.

Oct 21, 202635 views

Diffusion language model startup Inception released Mercury 3, its new flagship model, on October 20, 2026, according to SiliconANGLE.

How it works: instead of generating one token after another, a Mercury model drafts the whole response and refines every position in parallel over a few passes.
Speed: Inception measured 1,068 tokens/second on Nvidia GPUs, versus about 240 for Claude Haiku 5.5 and about 130 for GPT-6 Luna in the same comparison.
Benchmarks (company-reported): beat GPT-6 Luna and Claude Haiku 5.5 on 8 of 12 tests; 51.1 on SWE-Bench Atlas vs 33.3 (GPT-6 Luna) and 32.3 (Haiku 5.5); trailed Haiku 5.5 on DeepSWE (46.3 vs 51.6). No independent results had been published yet.
Context: 1M tokens, up from 260K in Mercury 2.5. Parameter count not disclosed (more than 40x Mercury 1).
Price: $0.20 per million input tokens and $0.70 per million output tokens, cached input 90% cheaper, with a launch promotion halving both rates.
Access: Inception's platform, OpenRouter, Baseten and Vercel; OpenAI-API-compatible code needs only minor changes.

Source: SiliconANGLE.