Mercury 2
Mercury 2
A diffusion model for text, and it is several times faster than anything built the usual way.
Updated
The read
Mercury generates text by refining a whole draft at once rather than one token after another, and the result is the highest measured output rate of any model in this index by a wide margin. For code completion and other places where the wait is the whole user experience, that architecture is a genuine advantage rather than a curiosity. Quality is good, not frontier, and the ecosystem around it is tiny.
Where it shines
- By far the highest measured output speed of anything ranked here
- Diffusion decoding refines a whole draft rather than one token at a time
- Strong at code completion specifically
Worth knowing first
- Quality sits below the frontier on anything demanding
- Unusual architecture, so tooling support is thin
Index score
73.2
Provisional. This score is read from the product's public capability, not from the one-prompt rebuild the fully reviewed entries went through, so treat it as a placing rather than a verdict.
- Rank
- 43 of 53
- Builds
- LLMs
- Category
- Coding
- Output
- Text and code over an API
- Pricing
- Paid only
- Best for
- Code completion where latency is the product
Scores are our own editorial judgement, weighted craft 55, speed 10, control 15 and value 20.
Also worth a look
Full index- 04Claude Sonnet 5Claude Sonnet 5ProvisionalThe coding workhorse: near the top on real software tasks, at a working price.Coding modelPaid only87.9
- 10MiniMax M3MiniMax M3ProvisionalOpen weights tuned for software work, and the fastest decoding of anything in its class.Coding modelFree tier84.5
- 18Step 3.7 FlashStep 3.7 FlashProvisionalAn Apache-licensed vision model built for coding agents, and it routes work rather than brute-forcing it.Coding modelFree tier81.6
