Inception releases Mercury 3, a fast diffusion language model aimed at agents
Announced · Published · By Contractor AI Shop
Inception launched Mercury 3, a diffusion language model that generates and refines many tokens in parallel instead of one at a time. Inception claims 1,068 tokens per second on NVIDIA GPUs, a 1M-token context and quality on par with GPT-6 Luna and Claude Haiku 5.5. Launch pricing is half off.
Key facts
- Inception lists standard pricing of $0.20 per million input and $0.70 per million output tokens, with a 50% launch discount to $0.10 and $0.35. Inception
- Inception says Mercury 3 is available on its API platform and through OpenRouter, Baseten and Vercel, and is OpenAI API compatible. Inception
- The Deep View reports Inception hasn't said how long the launch price lasts or how many parameters the model has. The Deep View (Jason Hiner)
- The Neuron's eight hands-on API runs found strong business math and coding but some sloppiness on exact constraints; it did not independently reproduce the 1,068 tokens/second figure. The Neuron (Corey Noles)
Our take: what it means for contractors
Speed is the pitch, and speed matters for agents that make many small calls, like voice answering and quick lookups. Inception offers it as a hosted API, not a download, so it isn't a local option. The Neuron's tests are a reminder to check its numbers and constraints before it touches a bid.
What we don't know yet
- Parameter count (Inception won't say, per The Deep View)
- How long the launch price lasts
- Independent benchmarks (the published scores are Inception's)
- Real throughput through third-party providers
Sources
- Inception · Diffusion at Scale: Introducing Mercury 3 · 2026-10-19 · Primary
- The Deep View (Jason Hiner) · Why Inception's Mercury 3 upends the AI status quo · 2026-10-19 · Secondary
- The Neuron (Corey Noles) · I Tested Mercury 3. Diffusion Language Models Just Got Serious. · 2026-10-20 · Secondary