Mercury 2.5 vs Mercury 3
Mercury 2.5 costs 2.3x less per output token than Mercury 3 ($0.15 vs $0.35 per million). Updated October 22, 2026; no benchmark has tested both models yet.
- Coding
- 1.6
- Overall
- No data
- Output $/M
- $0.15
- Coding
- No data
- Overall
- No data
- Output $/M
- $0.35
Mercury 2.5 or Mercury 3: which to pick
Only Mercury 2.5 has a coding score, 1.6 on the Vibe Coding Index; Mercury 3 has none, so the two cannot be ranked on coding.
Pick Mercury 3 for a larger context window (1M against 260K tokens).
No benchmark has tested both yet. Mercury 2.5 has results on Terminal-Bench 4 (Vals AI), Vibe Code Bench (Vals AI) and Code migration (Vals AI); Mercury 3 on Design Arena: websites, Design Arena: UI components and Design Arena: game dev. Compare them on price and context instead.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Only Mercury 2.5 has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
No call
Only Mercury 2.5 has a AA Intelligence Index result, 12.3.
Directional only
Web design
Design Arena: websites
No call
Only Mercury 3 has a Design Arena: websites result, 1242.
Directional only
Long documents
Listed context window
Mercury 3
Mercury 3 takes 1M tokens against 260K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Mercury 2.5
Mercury 2.5 $0.0001, Mercury 3 $0.0003 per turn.
List prices
At a glance
- Coding index
- Mercury 2.5 1.6 from 2 benchmarks, Mercury 3 not benchmarked.
- Widest category
- None yet: no category has been tested on both models.
- Widest single gap
- None yet: no benchmark has tested both models.
- Output price
- Mercury 2.5 $0.15 against Mercury 3 $0.35 per million output tokens, 2.3x apart.
- Context window
- Mercury 2.5 takes 260K tokens, Mercury 3 1M.
Benchmark by benchmark
Every board either model has a result on, a category to a fold; bars are each result on a 0 to 100 scale. Only one model is measured on each, so no row names a leader.
CodingDirectional onlySix of these boards feed the Vibe Coding Index9 boards measured
| Benchmark | Mercury 2.5 | Mercury 3 | Leader |
|---|---|---|---|
| Terminal-Bench 4 (Vals AI) | 0.0% | Not measured | Only Mercury 2.5 measured |
| Vibe Code Bench (Vals AI) | 3.7% | Not measured | Only Mercury 2.5 measured |
| Code migration (Vals AI) | 4.5% | Not measured | Only Mercury 2.5 measured |
| IOI (Vals AI) | 2.4% | Not measured | Only Mercury 2.5 measured |
| Terminal-Bench 2.1 (Vals AI) | 34.5% | Not measured | Only Mercury 2.5 measured |
| SkillsBench (Vals AI) | 18.1% | Not measured | Only Mercury 2.5 measured |
| CyberBench (Vals AI) | 45.5% | Not measured | Only Mercury 2.5 measured |
| SciCode | 38.5% | Not measured | Only Mercury 2.5 measured |
| ALE-Bench | 301.6 | Not measured | Only Mercury 2.5 measured |
| Neither model has a result on the other 30 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index1 board measured
| Benchmark | Mercury 2.5 | Mercury 3 | Leader |
|---|---|---|---|
| AA Intelligence Index | 12.3 | Not measured | Only Mercury 2.5 measured |
| Neither model has a result on the other 1 board here | |||
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index5 boards measured
| Benchmark | Mercury 2.5 | Mercury 3 | Leader |
|---|---|---|---|
| Finance Agent (Vals AI) | 18.9% | Not measured | Only Mercury 2.5 measured |
| Tax Agent Bench (Vals AI) | 2.5% | Not measured | Only Mercury 2.5 measured |
| Legal Research Bench (Vals AI) | 4.3% | Not measured | Only Mercury 2.5 measured |
| Excel Modeling Benchmark (Vals AI) | 8.8% | Not measured | Only Mercury 2.5 measured |
| Terminal-Bench-Science (Vals AI) | 0.0% | Not measured | Only Mercury 2.5 measured |
| Neither model has a result on the other 36 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index1 board measured
| Benchmark | Mercury 2.5 | Mercury 3 | Leader |
|---|---|---|---|
| Humanity's Last Exam (Artificial Analysis run) | 11.8% | Not measured | Only Mercury 2.5 measured |
| Neither model has a result on the other 21 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index1 board measured
| Benchmark | Mercury 2.5 | Mercury 3 | Leader |
|---|---|---|---|
| ProofBench | 3.0% | Not measured | Only Mercury 2.5 measured |
| Neither model has a result on the other 12 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index2 boards measured
| Benchmark | Mercury 2.5 | Mercury 3 | Leader |
|---|---|---|---|
| LegalBench (Vals AI) | 83.1% | Not measured | Only Mercury 2.5 measured |
| MedCode (Vals AI) | 31.3% | Not measured | Only Mercury 2.5 measured |
| Neither model has a result on the other 16 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index5 boards measured
| Benchmark | Mercury 2.5 | Mercury 3 | Leader |
|---|---|---|---|
| Design Arena: websites | Not measured | 1242 | Only Mercury 3 measured |
| Design Arena: UI components | Not measured | 1255 | Only Mercury 3 measured |
| Design Arena: game dev | Not measured | 1225 | Only Mercury 3 measured |
| Design Arena: data viz | Not measured | 1192 | Only Mercury 3 measured |
| Design Arena: all code categories | Not measured | 1237 | Only Mercury 3 measured |
| Neither model has a result on the other 4 boards here | |||
GeneralDirectional onlyPublic boards, shown as published, kept out of the index3 boards measured
| Benchmark | Mercury 2.5 | Mercury 3 | Leader |
|---|---|---|---|
| Public Benefits Bench (Vals AI) | 45.3% | Not measured | Only Mercury 2.5 measured |
| MedScribe (Vals AI) | 55.1% | Not measured | Only Mercury 2.5 measured |
| Long context reasoning (AA-LCR) | 71.7% | Not measured | Only Mercury 2.5 measured |
| Neither model has a result on the other 43 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Mercury 2.5 costs less
Repository review
50K input + 3K output tokens
Mercury 2.5 costs less
Agent loop
200K input + 10K output tokens
Mercury 2.5 costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Inception Mercury 2.5 | Inception Mercury 3 |
|---|---|---|
| Maker | Inception | Inception |
| Weights | No public weights listed | No public weights listed |
| Released | September 8, 2026 | October 16, 2026 |
| Context window | 260K | 1M |
| Input $/Mtok | $0.04 | $0.10 |
| Output $/Mtok | $0.15 | $0.35 |
| Coding index | 1.6 | Not benchmarked |
| Independent measurements behind it | 2 | 0 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Mercury 2.5 better than Mercury 3 for coding?
Only Mercury 2.5 has a coding score, 1.6 on the Vibe Coding Index; Mercury 3 has none, so the two cannot be ranked on coding. Pick Mercury 3 for a larger context window (1M against 260K tokens).
Which is cheaper, Mercury 2.5 or Mercury 3?
Mercury 2.5 costs $0.04 per million input tokens and $0.15 per million output tokens; Mercury 3 costs $0.10 and $0.35. Cost per solved task is known for neither of them.
Which has the bigger context window, Mercury 2.5 or Mercury 3?
Mercury 3, with 1M tokens against 260K.
Has any benchmark tested both Mercury 2.5 and Mercury 3?
No benchmark has tested both yet. Mercury 2.5 has results on Terminal-Bench 4 (Vals AI), Vibe Code Bench (Vals AI) and Code migration (Vals AI); Mercury 3 on Design Arena: websites, Design Arena: UI components and Design Arena: game dev. Compare them on price and context instead. Data as of October 22, 2026.