Inception
Fast diffusion language models for reasoning, coding, voice, support, and other applications where latency matters.
About Inception
Inception develops Mercury, a family of diffusion language models made for applications where response time shapes the user experience. Mercury 2 combines reasoning, tool use, structured output, and a 128K context window; Inception says its models run at more than 1,000 tokens per second on commercial NVIDIA GPUs. Mercury Edit 2 is a smaller, coding-focused model with a 32K context window for autocomplete and next-edit workflows. Both models use token-based pricing and expose an OpenAI-compatible API, which can reduce integration work for teams already using that interface. Inception also offers access through AWS Bedrock and Azure Foundry, while fine-tuning, private deployments, custom SLAs, and priority support are positioned for enterprise conversations. For buyers, the tradeoffs are concrete: teams gain unusually low generation latency and familiar API compatibility, but must evaluate diffusion-model behavior against their own workloads, account for separate input and output charges, and choose between the broader Mercury 2 feature set and the narrower coding focus of Mercury Edit 2.
Best for
Best use cases
A voice agent needs reasoning without long pauses between turns
A coding assistant runs tool-using subagents during rapid iteration
An editor requests autocomplete and next-edit suggestions as the user types
Compare before choosing
Mercury 2 has a 128K context window and broader reasoning features
Mercury Edit 2 has a 32K context window and a narrower code-editing focus
Both listed models share the same input, cached-input, and output token rates
Features
Realtime voice agents
Use Mercury 2 for responsive voice interactions where reasoning latency affects conversational flow.
Rapid coding workflows
Run coding iteration and workflow subagents with Mercury 2, including tool use and structured output.
Low-latency code editing
Use Mercury Edit 2 for autocomplete and next-edit suggestions in latency-sensitive developer tools.
Pros
- OpenAI API compatibility can reduce changes to an existing model integration.
- Mercury 2 combines a 128K context window with reasoning, tool use, and structured output.
- Token rates are published separately for input, cached input, and output usage.
Cons
- Usage-based charges apply separately to input, cached input, and output tokens.
- Mercury Edit 2 has a 32K context window and a narrower coding focus than Mercury 2.
- Private deployments, fine-tuning, custom SLAs, and priority support require a sales conversation.
Pricing Plans
Mercury 2
- 128K context window
- Reasoning
- Tool use
- Structured output
Mercury Edit 2
- 32K context window
- Autocomplete
- Next edit
* Pricing information is for reference only. Please refer to the latest announcement on Inception's official website.