Source: Unite.AI
Inception introduced Mercury 3 on October 19, 2026, describing the model as the most capable diffusion-based large language model released to date, with launch pricing of $0.10 per million input tokens and $0.35 per million output tokens.
The company said Mercury 3 outperforms GPT-6 Luna and Claude Haiku 5.5 across a range of benchmarks while running four to eight times faster at more than 1,000 tokens per second. The announcement lists a speed of 1,068 tokens per second on NVIDIA GPUs and a context window of one million tokens, up from 260,000 tokens for Mercury 2.5. List pricing stands at $0.20 per million input tokens and $0.70 per million output tokens, with a 50% discount applied at launch; cached input tokens are priced 90% below the input rate, listed at $0.02 per million tokens and $0.01 under the promotion.
Scaling Claims for Text Diffusion
Inception framed the release as evidence that text diffusion scales. According to the company, Mercury 3 is more than 40 times larger than Mercury 1 in total parameters, while intelligence across a range of standard benchmarks rose threefold from Mercury 1 (mini) to Mercury 3. Inception described Mercury 1 as the first proof point that diffusion LLMs could be built for commercial applications and said the open question in the AI community had been whether diffusion LLMs could scale to compete with autoregressive models on complex use cases. The company said Mercury 3 answers that question, and that the scaling laws that drove progress in autoregressive models hold for diffusion as well.
Early-Access Production Results
Inception said early-access partners are already running Mercury 3 in latency-sensitive applications including browser-use agents, interactive education, real-time search agents, and customer support. The company said it ran Mercury 3 inside its production web-agent harness against frontier models on 50 live Mind2Web navigation tasks, where Mercury 3 completed 50% of them, a completion count the company described as significantly higher than GPT-6.1-Sol’s, achieved with the same token volume and the same number of agent steps at a fraction of the per-token cost. The announcement also quotes Shuhao Zhang, cofounder of TinyFish AI, in its browser-use section.
Michael Tong, co-founder and CTO of General Learning, known as RevisionDojo, said Mercury 3 let the tutoring company give every student instant feedback “without giving up reasoning,” with hints, marked answers, and definitions returning in about a second at quality he described as on par with or better than the models the company previously ran. He called the switch a drop-in swap for the company’s existing stack.
Kuan Zou, founder and CEO of Octen AI, said that on smart glasses “an answer that takes five seconds feels broken.” In Octen’s Sense API tests, he reported, Mercury 3 delivered a median generation time of 0.9 seconds and a median end-to-end response time of 2.2 seconds, which he described as the best balance of speed and quality among the fast models the company tested.
Michael Gribben, a member of technical staff at Lorikeet, which builds customer concierges across voice, chat, and email for complex and regulated businesses, said Mercury 3 beat other models on latency in early testing while also performing higher on evaluations and in production.
The Wider Mercury Portfolio
Mercury 3 is the flagship of a product portfolio Inception describes as built for real-time AI. Mercury Voice is customized for the low-latency requirements of voice applications, with a focus on minimizing time to first answer token, and Inception said it already powers agents at Audivi AI, Altur, and OpenCall. The company’s models page lists Mercury Voice with time-to-first-token under 170 milliseconds, priced at $0.40 per million input tokens and $1.50 per million output tokens, with promotional rates of $0.20 and $0.75 displayed.
Mercury Decide is a decision model built for high-volume classification. It returns answers with calibrated probabilities and can escalate uncertain cases to a human reviewer. Inception said Mercury Decide outperforms Jev on benchmarks including JevBench and LegalBench. The models page prices it at $0.04 per million input tokens, discounted to $0.02 under the displayed promotion, and lists output tokens at $0.
Inception’s blog listing shows Mercury Voice was introduced on September 29, 2026, and that an October 15, 2026 post announced Mercury 2 was named to TIME’s Best Inventions of 2026.
API Access and the Mercury 2 Sunset
Mercury 3 is available through the Inception API Platform and through OpenRouter, Baseten, and Vercel, and is OpenAI API compatible. The company’s quick-start documentation identifies the model as mercury-3 served from the api.inceptionlabs.ai/v1 chat completions endpoint, with request parameters including reasoning_effort and temperature, and offers inceptionai SDKs for Python and TypeScript. The documentation also shows the API working through AISuite, LiteLLM, LangChain, the OpenAI client, and the Vercel AI SDK. New platform accounts include 100 million free tokens, with no payment details required.
For enterprise deployments, Inception lists dedicated capacity, autoscaling, compliance controls, and configurable data retention, and said it will partner with enterprise customers on workload fit, evaluation design, and performance under their serving constraints.
Alongside the launch, the models page notes that Mercury 2 and Mercury 2.5 are no longer available to new users. Existing customers with usage before October 19, 2026 will keep access until November 15, 2026, and enterprise customers remain supported under their existing agreements.
