Inception

Inception announced Mercury 2.5 on September 8, 2026 in a first-party blog post authored by CEO Stefano Ermon, positioning it as the company's most capable production diffusion large language model to date. The post claims a 40% intelligence increase over Mercury 2, with Mercury 2.5 described as "the largest diffusion The same blog outlines production use cases driving Mercury 2.5's design, including search agents and RAG pipelines where multiple model calls must fit inside a single user interaction, and voice agents where latency directly shapes caller experience. The post credits customer feedback and production failure cases, rat

Inception announced the launch of the Inception API in a first-party blog post by Chief of Staff Sawyer Birnbaum, providing programmatic access to Inception's commercial-scale diffusion large language models. The API debuted with support for Mercury Coder Small, a coding-focused model the post claims runs more than 5x The post describes the Inception API as OpenAI-compatible, exposing chat/completions for conversational interactions and fim/completions for fill-in-the-middle coding workflows. Mercury Coder Small's documented capabilities include a 32k token context window, 16k maximum output tokens, support for languages such as Pyt

Inception's first-party blog introduced Mercury 2, the prior-generation diffusion reasoning model authored by CEO Stefano Ermon and described as the world's fastest reasoning language model at launch. Mercury 2 generates responses through parallel refinement rather than sequential autoregressive decoding, producing mul The Mercury 2 launch post emphasizes latency-sensitive production workloads where Mercury 2 excels, framing autoregressive decoding as a compounding bottleneck in agent loops, retrieval pipelines, and extraction jobs. It positions diffusion-based reasoning as enabling reasoning-grade quality inside real-time latency bu

An aggregator timeline lists an Inception: Mercury 2.5 API launch on September 8, 2026, with a 260k token context window, providing independent corroboration of the release date and context length. The entry sits among other 2026 launches including DeepSeek V4.1 Flash, Gemini 3.8 Flash, and GPT-6 Astra on the OpenRoute The list is auto-refreshed from OpenRouter model metadata rather than sourced directly from Inception, so it functions as a timestamp and surface-area confirmation rather than a detailed technical reference. The Mercury 2.5 entry appears at position seven in the September 2026 portion of the timeline.

An independent article on shattered.io, dated September 8, 2026, covers Inception's Mercury 2.5 launch and corroborates the first-party announcement's headline performance claim of more than 1,100 tokens per second in production deployments. The piece frames Mercury 2.5 as the fastest reasoning model in live production The shattered.io coverage adds detail not present in the primary blog, reporting that Mercury 2.5 shipped alongside two companion previews: Mercury Voice, a voice interface layered on the same diffusion backbone, and Mercury Router, a routing product. The article positions Mercury 2.5 as a step up in what Inception cal