Text and chat · Inception
Mercury
Diffusion language models that write many tokens at once: very fast replies, with reasoning and tools, at a low price.
Anonymized Low costWhat sets it apart
Mercury models are diffusion language models: they refine many tokens in parallel instead of writing one at a time. That makes them several times faster than usual models, which suits voice agents and coding tools, while their intelligence scores stay modest.
- Diffusion decoding, not token by token
- Over 1,000 tokens per second, per Inception
- Tunable reasoning and parallel tool calls
Recommended for: get quick answers for very little.
Strengths and limits
Where it shines
- Very fast output: over 1,000 tokens per second, according to Inception
- Low price, tool calling, adjustable reasoning
Think twice when
- Anonymized, not private: Inception's retention policy applies
- Text only, no image reading
- Mercury 2.5: 260K context
- Diffusion model, not word by word
Use cases
Instant replies in a voice assistant
You are a phone assistant for a dental clinic. A caller asks to move Thursday’s 3 pm appointment to Friday morning. Answer in two short sentences and ask one clarifying question.
Very low latency keeps a spoken exchange natural.
Fast code edits in an editor
Rename the function getUser to fetchUser in this file and update every call. Return only the changed code.
Short, repetitive edits benefit from raw speed more than from deep reasoning.
Against its rivals
- GeminiGemini Flash models are fast and score higher, with image input.#5 of 90 on Artificial Analysis Intelligence Index (53 points)
- GPT OSSgpt-oss is open and reasons deeper, but generates far more slowly.#79 of 90 on Artificial Analysis Intelligence Index (12 points)
- Mistral SmallMistral Small is similarly cheap and reads images, without diffusion speed.#87 of 90 on Artificial Analysis Intelligence Index (8 points)
Version history
Each version, dated, with what it changed over the one before.
Hollow points are earlier versions no longer offered in Sub Rosa.
See the numbers
| Version | Released | points |
|---|---|---|
| Mercury 2 | 2026-02-24 | 14 |
| Mercury 2.5 | 2026-10-01 | 12 |
- Mercury 2
- Diffusion-based reasoning model with tool use and schema-aligned JSON, 128K context
- Under 300 ms time to first token for voice use, per Inception
- Mercury 2.5Latest
- Context grows from 128K to 260K tokens
- Tunable reasoning controls and parallel tool calls
- About 1,107 tokens per second and 40% more intelligence than Mercury 2, per Inception
Benchmarks
How to read these scores: what a benchmark tells you.
Artificial Analysis Intelligence Index: #76 of 90 models measured in the catalog. One score averaging about ten independent tests of reasoning, knowledge, math, coding and agent work, all run the same way for every model. Higher is better. Each model is listed at its highest reasoning effort. Two or three points apart is a close call; ten points is a different class of model.
See the numbers
| Model | points |
|---|---|
| 1. Claude Opus 5.5 | 58 |
| 2. Claude Sonnet 5.5 | 56 |
| 3. Claude Fable 5.1 | 53 |
| 4. GPT-6 Astra | 53 |
| 5. Gemini 4 Argon | 53 |
| 6. GPT-6.1 Sol | 52 |
| 7. Muse Wonder | 52 |
| 8. Claude Opus 5 | 51 |
| 9. Claude Fable 5 | 50 |
| 10. GPT-6 Sol | 48 |
| 76. Mercury 2 | 14 |
| 82. Mercury 2.5 | 12 |
All 2 models in this family
| Model | Use | Privacy | Price |
|---|---|---|---|
Mercury 2.5mercury-2-5 | Text | Anonymized | $0.422 per million tokens written$0.000295 a page written |
Mercury 2mercury-2 | Text | Anonymized | $0.422 per million tokens written$0.000295 a page written |
Specifications
| Specification | Mercury 2.5 | Mercury 2 |
|---|---|---|
| Released | October 1, 2026 | February 24, 2026 |
| Weights | Closed | Closed |
| Accepts | text | text |
| Produces | text | text |
| Context | 260K tokens | 128K tokens |
| Longest answer | 66K tokens | 50K tokens |
| Abilities | Uses Sub Rosa’s tools, Thinks before answering | Uses Sub Rosa’s tools, Thinks before answering |
| Price | $0.422 per million tokens written | $0.422 per million tokens written |
| Privacy | Anonymized | Anonymized |
Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.
Sources
Everything on this page comes from these pages, read on October 16, 2026 or as dated on each score.