Mercury Coder
by Inception Labs · United StatesA diffusion language model rather than an autoregressive one — it denoises whole blocks of tokens at once, which is why it is so fast.
Specification
The numbers
- Maker
- Inception Labs
- Released
- 2025-02
- Parameters
- Not reported
- Architecture
- Not reported
- Context window
- 32,768 tokens
- Max output
- 16,384 tokens
- Input
- Text
- Output
- Text
- Reasoning
- No
- Tool calling
- Yes
- Knowledge cutoff
- Not reported
- Licence
- Proprietary
- Availability
- Not reported
- Weights
- Not released
Cost
Price per million tokens
- Input
- $0.25 / M tokens
- Output
- $1.00 / M tokens
- Blended 3:1
- $0.438
Standard first-party API rate, excluding batch discounts. Reasoning models bill thinking tokens as output, so cost per answer can far exceed the cost per token suggests.
Published scores
Benchmarks
Figures published by Inception Labs or taken from a public leaderboard. Row last checked 2026-08.
| Category | Score | Rank |
|---|---|---|
| ReasoningMulti-step logic on problems that cannot be looked up | Not reported | — |
| MathsCompetition mathematics, graded on the final answer | Not reported | — |
| CodingHumanEval | 90.0% | 9 of 46 reporting the same tests |
| KnowledgeBreadth of factual recall under exam conditions | Not reported | — |
| MultimodalReading charts, diagrams and photographs | Not reported | — |
| Instruction followingObeying an exact, checkable format | Not reported | — |
| Human preferenceWhich answer people pick, blind | Not reported | — |
A category averages every benchmark in it that Mercury Coder reports. The rank counts only models that report the same tests, so it never compares an average over three benchmarks against an average over one.
Every reported test
| MMLU-ProKnowledge | Not reported | No figure published |
|---|---|---|
| GPQA DiamondReasoning | Not reported | No figure published |
| AIME 2025Maths | Not reported | No figure published |
| MATH-500Maths | Not reported | No figure published |
| SWE-bench VerifiedCoding | Not reported | No figure published |
| SWE-bench ProCoding | Not reported | No figure published |
| Terminal-Bench 2.1Coding | Not reported | No figure published |
| Frontier-Bench v0.1Reasoning | Not reported | No figure published |
| Terminal-Bench 4.0Coding | Not reported | No figure published |
| LiveCodeBenchCoding | Not reported | No figure published |
| HumanEvalCoding | Rank 9 of 46 models reporting | |
| MMMUMultimodal | Not reported | No figure published |
| IFEvalInstruction following | Not reported | No figure published |
| LMArena EloHuman preference | Not reported | No figure published |
Where these numbers come from
Every score on this page is a published figure, taken from the model's own card, system card, technical report or release post, or from a public leaderboard. CorX Labs did not run these evaluations. Most are self-reported by the lab that built the model, which means they were produced under that lab's own choice of prompt, scaffold and number of attempts — so treat them as a starting point for a shortlist, not as a settled ranking.
A score someone other than the model's maker measured is marked Independent and names its measurer. Those are the stronger numbers on this page — an outside harness has no reason to flatter anyone — and there are not many of them.
Where a figure has not been published, the cell reads Not reported rather than an estimate. Nothing here is inferred, interpolated or guessed. Each model records the month its row was last checked. Full method and caveats.