Key Takeaways
- • Inception announced Mercury 3 on October 19, 2026, which it describes as the most capable diffusion large language model (dLLM) released to date, claiming quality on par with or better than GPT-6 Luna and Claude Haiku 5.5 while running above 1,000 tokens per second.
- • According to Inception, Mercury 3 runs 4 to 8 times faster than GPT-6 Luna and Claude Haiku 5.5 on a range of benchmarks and is more than 40 times the size of Mercury 1 at comparable speeds.
- • Mercury 3 offers a 1-million-token context window, up from 260,000 tokens in Mercury 2.5, and is priced at $0.20 per million input tokens and $0.70 per million output tokens, with a launch discount to $0.10 and $0.35.
- • The model is available through the Inception API, Baseten, OpenRouter, and Vercel, and is OpenAI API compatible, with enterprise options including dedicated capacity, autoscaling, compliance controls, and configurable data retention.
- • Inception is backed by Menlo Ventures, Mayfield, Innovation Endeavors, M12, Snowflake Ventures, Databricks Ventures, and individual angels including Andrew Ng, Andrej Karpathy, and Eric Schmidt.
Summary
Redwood City, California-based Inception announced Mercury 3 on October 19, 2026, positioning it as the most capable diffusion large language model released to date. According to the company's press release distributed via Business Wire, Mercury 3 matches or exceeds frontier speed-optimized models GPT-6 Luna and Claude Haiku 5.5 across a broad range of standard benchmarks while sustaining inference speeds above 1,000 tokens per second on widely available NVIDIA GPUs. Inception characterizes the release as demonstrating that diffusion models can scale in size and capability without sacrificing speed.
Mercury 3 is Inception's fourth model since Mercury 1 launched in June 2025. Mercury 2 introduced reasoning to diffusion models, and Mercury 2.5 added tunable reasoning and a 260,000-token context window. Mercury 3 expands the context window to 1 million tokens and is more than 40 times the size of Mercury 1, according to the company. Stefano Ermon, co-founder of Inception, said the limits to scaling diffusion models are compute and data rather than fundamental algorithmic barriers, and that the company has already begun training its next model.
The company contrasts diffusion models with traditional autoregressive LLMs, which generate text one token at a time. Diffusion models generate multiple tokens in parallel, which Inception argues is better suited to latency-sensitive agentic workloads, where delays compound across repeated model calls. Early partners cited in the release include General Learning, operator of RevisionDojo, whose CTO Michael Tong said Mercury 3 returns feedback in about a second with quality equal to or better than prior models and worked as a drop-in replacement. Other early use cases include browser-use agents, interactive education, real-time search agents, and customer support.
Mercury 3 is priced at $0.20 per million input tokens and $0.70 per million output tokens, with a launch discount to $0.10 and $0.35 respectively and a 90% discount on cached input tokens. The model is available via the Inception API, Baseten, OpenRouter, and Vercel, and its OpenAI API compatibility allows teams to switch without rewriting code. Enterprise deployments include dedicated capacity, autoscaling, compliance controls, and configurable data retention. Inception is backed by Menlo Ventures, Mayfield, Innovation Endeavors, M12 (Microsoft's venture arm), Snowflake Ventures, Databricks Ventures, and individual investors including Andrew Ng, Andrej Karpathy, and Eric Schmidt. The benchmark claims are company-reported and were not independently verified in the source material.
Mercury 3 is Inception's fourth model since Mercury 1 launched in June 2025. Mercury 2 introduced reasoning to diffusion models, and Mercury 2.5 added tunable reasoning and a 260,000-token context window. Mercury 3 expands the context window to 1 million tokens and is more than 40 times the size of Mercury 1, according to the company. Stefano Ermon, co-founder of Inception, said the limits to scaling diffusion models are compute and data rather than fundamental algorithmic barriers, and that the company has already begun training its next model.
The company contrasts diffusion models with traditional autoregressive LLMs, which generate text one token at a time. Diffusion models generate multiple tokens in parallel, which Inception argues is better suited to latency-sensitive agentic workloads, where delays compound across repeated model calls. Early partners cited in the release include General Learning, operator of RevisionDojo, whose CTO Michael Tong said Mercury 3 returns feedback in about a second with quality equal to or better than prior models and worked as a drop-in replacement. Other early use cases include browser-use agents, interactive education, real-time search agents, and customer support.
Mercury 3 is priced at $0.20 per million input tokens and $0.70 per million output tokens, with a launch discount to $0.10 and $0.35 respectively and a 90% discount on cached input tokens. The model is available via the Inception API, Baseten, OpenRouter, and Vercel, and its OpenAI API compatibility allows teams to switch without rewriting code. Enterprise deployments include dedicated capacity, autoscaling, compliance controls, and configurable data retention. Inception is backed by Menlo Ventures, Mayfield, Innovation Endeavors, M12 (Microsoft's venture arm), Snowflake Ventures, Databricks Ventures, and individual investors including Andrew Ng, Andrej Karpathy, and Eric Schmidt. The benchmark claims are company-reported and were not independently verified in the source material.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.