Mercury 2.5
Mercury 2.5
A diffusion model for text that writes over a thousand tokens a second.
Updated
The read
Mercury generates text by refining a whole draft at once rather than one token after another, and version 2.5, released in September, pushes that to roughly 1,100 tokens a second. Inception claims a large jump in quality over Mercury 2, though it has published no benchmark table to back that up, and the independent index still places it among the small models. Where latency is the product, as in code completion, nothing else here is as quick.
Where it shines
- Around 1,100 tokens a second, far beyond conventional models
- Launch pricing of twenty cents in and seventy-five out per million tokens
- 260,000 tokens of context
Worth knowing first
- No published benchmark table for this release
- Reasoning depth is well below the conventional models it outpaces
Index score
75.2
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
- 62 of 73
- Builds
- LLMs
- Category
- Coding
- Output
- Text and code over an API
- Pricing
- Paid only
- Best for
- Code completion and edits 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- 02Claude Sonnet 5.5Claude Sonnet 5.5ProvisionalNearly the strongest model in the index, at a mid-range price. Sonnet stopped being the compromise.Coding modelPaid only91.5
- 24MiniMax M3MiniMax M3ProvisionalOpen weights tuned for software work, and the fastest decoding of anything in its class.Coding modelFree tier84.5
- 33Claude Sonnet 5Claude Sonnet 5ProvisionalSonnet 5.5 costs exactly the same and is far stronger. There is little reason to stay.Coding modelPaid only82.5
