Mercury 2
Mercury 2The fastest reasoning LLM, powered by diffusion.
Model details
Example usage
Mercury 2 is Inception's diffusion-based reasoning model, built to make production AI feel instant.
Instead of decoding one token at a time like standard autoregressive LLMs, it generates responses through parallel refinement — converging on an answer in a small number of steps rather than a long left-to-right chain.
It still supports the things production apps need — tunable reasoning, 128K context, native tool use, and schema-aligned JSON output — and it's OpenAI API compatible, so it drops into an existing stack without a rewrite. It's best suited for latency-sensitive workloads: coding assistants, agentic loops, voice agents, and search/RAG pipelines.
Here's a quick example calling it via the API:
1from inceptionai import Inception
2
3client = Inception(
4 api_key=os.environ.get("INCEPTION_API_KEY"), # defaults to this env var; can be omitted
5)
6
7chat_completion = client.chat.completions.create(
8 model="mercury-2",
9 messages=[
10 {
11 "role": "user",
12 "content": "What is a diffusion language model?"
13 }
14 ],
15 max_tokens=256,
16 temperature=0.75,
17)
18print(chat_completion)
191{
2 "id": "chatcmpl-7a2b3c4d5e",
3 "object": "chat.completion",
4 "created": 1745798400,
5 "model": "mercury-2",
6 "choices": [
7 {
8 "index": 0,
9 "finish_reason": "stop",
10 "message": {
11 "role": "assistant",
12 "content": "A diffusion language model is a type of language model that uses diffusion to generate text."
13 }
14 }
15 ],
16 "usage": {
17 "prompt_tokens": 12,
18 "completion_tokens": 8,
19 "total_tokens": 20,
20 "reasoning_tokens": 0,
21 "cached_input_tokens": 0
22 }
23}
