Mercury 3

Inception's flagship diffusion LLM: ~1,068 tokens/sec, 1M-token context, $0.20/$0.70 per million tokens.

Inception · Frontier/foundation models · model

Published in Feed

Overview

Inception introduced Mercury 3 on October 19, 2026, its flagship diffusion-based language model, which it says is over 40x larger than Mercury 1 in total parameters. It generates about 1,068 tokens per second on NVIDIA GPUs, which Inception puts at 4-8x faster than comparable models, and extends context to 1M tokens from 260K in Mercury 2.5. Inception claims it matches or beats GPT-6 Luna and Claude Haiku 5.5 across several benchmarks and completed 50% of 50 live Mind2Web navigation tasks (ahead of GPT-6.1 Sol in the same test); the post gives no per-benchmark scores. It is OpenAI-API compatible and available on the Inception API Platform, OpenRouter, Baseten and Vercel.

This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.

Key features

  • About 1,068 tokens/second on NVIDIA GPUs (vendor-reported)
  • 1M-token context window, up from 260K in Mercury 2.5
  • $0.20 input / $0.70 output per million tokens; launch discount to $0.10 / $0.35; cached input 90% cheaper
  • OpenAI-compatible API; on Inception Platform, OpenRouter, Baseten and Vercel AI Gateway
  • 100 million free tokens on the Inception API platform

Use cases

  • Browser and web-navigation agents (early partner TinyFish AI)
  • High-throughput customer-support and search workloads (Lorikeet, Octen AI)
  • Latency-sensitive agent loops where tokens/sec dominates cost

Pricing

$0.20 per 1M input and $0.70 per 1M output tokens; launch discount to $0.10 / $0.35; cached input tokens 90% cheaper; 100M free tokens on the API platform; enterprise plans add dedicated capacity.

Pricing reflects the source record and may change. Confirm current terms with the provider.

Dates and sources

Released
19 Oct 2026
Record updated
20 Oct 2026
Record last checked
20 Oct 2026

Explore the Feed

Other entries filed under frontier/foundation models; inclusion does not imply equivalent features.

  • Muse Wonder

    Meta's most capable Muse models, with active memory and long-goal focus for agents, 1M context and $4/$20 per million tokens.

  • Rho-1

    Reka's 19B omni-reasoning model that handles text, images, video and robot actions in one network, in research preview.

  • Amazon Nova 3

    Amazon's third-generation Nova frontier model family, unveiled at an AWS keynote with a 1M-token context window and a dedicated agentic tool-use variant.

Browse the Feed