Nemotron 3 Ultra vs MiMo-V2.5-Pro
MiMo-V2.5-Pro leads Nemotron 3 Ultra on coding, 31.6 to 30.0 on the Vibe Coding Index, and MiMo-V2.5-Pro costs 2.6x less per output token ($0.9152 vs $2.40 per million). Updated October 22, 2026; 62 benchmarks have a result for both.
Nemotron 3 Ultra or MiMo-V2.5-Pro: which to pick
Nemotron 3 Ultra and MiMo-V2.5-Pro are level for coding, 30.0 against 31.6 on the Vibe Coding Index, so price and context decide.
Pick Nemotron 3 Ultra for work measured by LiveCodeBench (Vals AI) or GPQA Diamond, where it leads. Pick MiMo-V2.5-Pro for output tokens 2.6x cheaper ($0.9152 vs $2.40 per million) and a larger context window (1.05M against 262K tokens).
- Shared results
- 62
- Nemotron 3 Ultra only
- 9
- MiMo-V2.5-Pro only
- 9
- Like-for-like categories
- 8 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
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
Level: 30.0 against 31.6, inside 2 points.
3 and 4 benchmarks behind the scores
Intelligence
AA Intelligence Index
MiMo-V2.5-Pro
MiMo-V2.5-Pro leads, 26.0 against 22.9.
Same board, both measured
Agentic work
AA Agentic Index
MiMo-V2.5-Pro
MiMo-V2.5-Pro leads, 21.3 against 20.1.
Same board, both measured
Reasoning
GPQA Diamond
Nemotron 3 Ultra
Nemotron 3 Ultra leads, 86.1% against 82.6%.
Same board, both measured
Knowledge
MMLU Pro
Nemotron 3 Ultra
Nemotron 3 Ultra leads, 85.8% against 84.6%.
Same board, both measured
Web design
Design Arena: websites
MiMo-V2.5-Pro
MiMo-V2.5-Pro leads, 1266 against 1145.
Same board, both measured
Long documents
Listed context window
MiMo-V2.5-Pro
MiMo-V2.5-Pro takes 1.05M tokens against 262K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
MiMo-V2.5-Pro
Nemotron 3 Ultra $0.0018, MiMo-V2.5-Pro $0.0009 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only MiMo-V2.5-Pro has the number, $0.03.
Directional only
At a glance
- Coding index
- Nemotron 3 Ultra 30.0 from 3 benchmarks, MiMo-V2.5-Pro 31.6 from 4 benchmarks.
- Widest category
- Web design, 13.6 points apart on average over 8 shared benchmarks.
- Widest single gap
- Vibe Code Bench (Vals AI): Nemotron 3 Ultra 7.6% against MiMo-V2.5-Pro 34.1%, 26.5 points apart on a 0 to 100 scale.
- Output price
- Nemotron 3 Ultra $2.40 against MiMo-V2.5-Pro $0.9152 per million output tokens, 2.6x apart.
- Context window
- Nemotron 3 Ultra takes 262K tokens, MiMo-V2.5-Pro 1.05M.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Nemotron 3 Ultra | Bars | MiMo-V2.5-Pro | Ties |
|---|---|---|---|---|
| Coding | 1 | 9 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 5 | ||
| Reasoning | 2 | 4 | ||
| Math | 0 | 3 | ||
| Knowledge | 4 | 5 | ||
| Web design | 0 | 8 | ||
| General | 4 | 16 | ||
| All shared boards | 11 | 51 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index19
| Benchmark | Nemotron 3 Ultra | MiMo-V2.5-Pro | Leader |
|---|---|---|---|
| LMArena WebDev | Not measured | 1480 | Only MiMo-V2.5-Pro measured |
| AA Coding Index | 49.3 | 60.2 | MiMo-V2.5-Pro |
| LiveCodeBench (Vals AI) | 86.0% | 81.3% | Nemotron 3 Ultra |
| SWE-bench Verified (Vals AI) | 69.0% | 74.0% | MiMo-V2.5-Pro |
| Terminal-Bench 4 (Vals AI) | Not measured | 0.5% | Only MiMo-V2.5-Pro measured |
| Vibe Code Bench (Vals AI) | 7.6% | 34.1% | MiMo-V2.5-Pro |
| Code migration (Vals AI) | 4.9% | 21.6% | MiMo-V2.5-Pro |
| ProgramBench (Vals AI) | 0.0% | Not measured | Only Nemotron 3 Ultra measured |
| Terminal-Bench 2.1 (Vals AI) | 50.9% | 57.3% | MiMo-V2.5-Pro |
| SciCode | 40.3% | 50.6% | MiMo-V2.5-Pro |
| Terminal-Bench Hard (Artificial Analysis run) | 36.4% | 43.2% | MiMo-V2.5-Pro |
| FrontierCode | 13.6% | Not measured | Only Nemotron 3 Ultra measured |
| WeirdML | 43.5% | Not measured | Only Nemotron 3 Ultra measured |
| ALE-Bench | Not measured | 899.8 | Only MiMo-V2.5-Pro measured |
| LMArena Text: coding | 1468 | 1501 | MiMo-V2.5-Pro |
| LMArena Text: software and IT | 1468 | 1495 | MiMo-V2.5-Pro |
| LMArena WebDev: React | Not measured | 1469 | Only MiMo-V2.5-Pro measured |
| LMArena WebDev: HTML | Not measured | 1473 | Only MiMo-V2.5-Pro measured |
| Neither model has a result on the other 25 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Nemotron 3 Ultra | MiMo-V2.5-Pro | Leader |
|---|---|---|---|
| AA Intelligence Index | 22.9 | 26.0 | MiMo-V2.5-Pro |
| Epoch Capabilities Index | 146.0 | Not measured | Only Nemotron 3 Ultra measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index05
| Benchmark | Nemotron 3 Ultra | MiMo-V2.5-Pro | Leader |
|---|---|---|---|
| AA Agentic Index | 20.1 | 21.3 | MiMo-V2.5-Pro |
| Finance Agent (Vals AI) | 37.7% | 41.5% | MiMo-V2.5-Pro |
| Tax Agent Bench (Vals AI) | Not measured | 11.5% | Only MiMo-V2.5-Pro measured |
| Legal Research Bench (Vals AI) | 15.4% | 15.9% | MiMo-V2.5-Pro |
| Excel Modeling Benchmark (Vals AI) | 31.7% | 55.2% | MiMo-V2.5-Pro |
| τ²-Bench Telecom | 83.3% | 94.2% | MiMo-V2.5-Pro |
| APEX-Agents | 22.7% | Not measured | Only Nemotron 3 Ultra measured |
| MCP Atlas | 63.1% | Not measured | Only Nemotron 3 Ultra measured |
| Neither model has a result on the other 34 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index24
| Benchmark | Nemotron 3 Ultra | MiMo-V2.5-Pro | Leader |
|---|---|---|---|
| GPQA Diamond | 86.1% | 82.6% | Nemotron 3 Ultra |
| GPQA Diamond (Artificial Analysis run) | 86.7% | 86.6% | Nemotron 3 Ultra |
| Humanity's Last Exam (Artificial Analysis run) | 28.4% | 35.7% | MiMo-V2.5-Pro |
| GPQA Diamond (Epoch AI run) | 85.4% | Not measured | Only Nemotron 3 Ultra measured |
| Mystery game puzzles | 20.0% | Not measured | Only Nemotron 3 Ultra measured |
| LMArena Text: hard prompts | 1452 | 1486 | MiMo-V2.5-Pro |
| LMArena Text: expert | 1472 | 1501 | MiMo-V2.5-Pro |
| LMArena Text: hard prompts in English | 1453 | 1494 | MiMo-V2.5-Pro |
| Neither model has a result on the other 15 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | Nemotron 3 Ultra | MiMo-V2.5-Pro | Leader |
|---|---|---|---|
| OTIS Mock AIME | 86.7% | Not measured | Only Nemotron 3 Ultra measured |
| ProofBench | 2.0% | 22.0% | MiMo-V2.5-Pro |
| LMArena Text: math | 1457 | 1481 | MiMo-V2.5-Pro |
| LMArena Text: mathematical work | 1461 | 1491 | MiMo-V2.5-Pro |
| Neither model has a result on the other 10 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index45
| Benchmark | Nemotron 3 Ultra | MiMo-V2.5-Pro | Leader |
|---|---|---|---|
| MMLU Pro | 85.8% | 84.6% | Nemotron 3 Ultra |
| LegalBench (Vals AI) | 82.1% | 77.1% | Nemotron 3 Ultra |
| CorpFin (Vals AI) | 65.5% | 61.4% | Nemotron 3 Ultra |
| TaxEval (Vals AI) | 73.1% | 73.8% | MiMo-V2.5-Pro |
| MedCode (Vals AI) | 38.6% | 32.5% | Nemotron 3 Ultra |
| LMArena Text: science | 1465 | 1482 | MiMo-V2.5-Pro |
| LMArena Text: medicine and healthcare | 1456 | 1473 | MiMo-V2.5-Pro |
| LMArena Text: legal and government | 1452 | 1474 | MiMo-V2.5-Pro |
| LMArena Text: business and finance | 1434 | 1466 | MiMo-V2.5-Pro |
| Neither model has a result on the other 11 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index08
| Benchmark | Nemotron 3 Ultra | MiMo-V2.5-Pro | Leader |
|---|---|---|---|
| Design Arena: websites | 1145 | 1266 | MiMo-V2.5-Pro |
| Design Arena: UI components | 1135 | 1246 | MiMo-V2.5-Pro |
| Design Arena: game dev | 1138 | 1261 | MiMo-V2.5-Pro |
| Design Arena: data viz | 1141 | 1257 | MiMo-V2.5-Pro |
| Design Arena: 3D | 1131 | 1243 | MiMo-V2.5-Pro |
| Design Arena: all code categories | 1146 | 1265 | MiMo-V2.5-Pro |
| Design Arena: SVG | 1069 | 1182 | MiMo-V2.5-Pro |
| Design Arena: ASCII art | 1100 | 1156 | MiMo-V2.5-Pro |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index416
| Benchmark | Nemotron 3 Ultra | MiMo-V2.5-Pro | Leader |
|---|---|---|---|
| MedScribe (Vals AI) | Not measured | 83.7% | Only MiMo-V2.5-Pro measured |
| IFBench | 81.4% | 79.9% | Nemotron 3 Ultra |
| Long context reasoning (AA-LCR) | 79.3% | 79.7% | MiMo-V2.5-Pro |
| DTBench | 90.1% | 84.5% | Nemotron 3 Ultra |
| LMCA | 36.9% | 29.5% | Nemotron 3 Ultra |
| LMArena Text | 1445 | 1464 | MiMo-V2.5-Pro |
| LMArena Text: instruction following | 1418 | 1475 | MiMo-V2.5-Pro |
| LMArena Text: creative writing | 1399 | 1438 | MiMo-V2.5-Pro |
| LMArena Text: English | 1455 | 1477 | MiMo-V2.5-Pro |
| LMArena Text: non-English | 1427 | 1448 | MiMo-V2.5-Pro |
| LMArena Text: Chinese | 1497 | 1507 | MiMo-V2.5-Pro |
| LMArena Text: German | 1472 | 1459 | Nemotron 3 Ultra |
| LMArena Text: Japanese | Not measured | 1410 | Only MiMo-V2.5-Pro measured |
| LMArena Text: Korean | 1384 | 1435 | MiMo-V2.5-Pro |
| LMArena Text: French | 1472 | 1487 | MiMo-V2.5-Pro |
| LMArena Text: Spanish | 1455 | 1469 | MiMo-V2.5-Pro |
| LMArena Text: Russian | 1417 | 1449 | MiMo-V2.5-Pro |
| LMArena Text: Polish | 1385 | 1464 | MiMo-V2.5-Pro |
| LMArena Text: multi-turn | 1411 | 1476 | MiMo-V2.5-Pro |
| LMArena Text: longer queries | 1437 | 1481 | MiMo-V2.5-Pro |
| LMArena Text: writing and language | 1406 | 1451 | MiMo-V2.5-Pro |
| LMArena Text: entertainment and media | 1397 | 1438 | MiMo-V2.5-Pro |
| EQ-Bench 4 | Not measured | 1208 | Only MiMo-V2.5-Pro measured |
| Neither model has a result on the other 25 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Nemotron 3 Ultra | MiMo-V2.5-Pro | Leader |
|---|---|---|---|
| Neither model has a result on 5 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
MiMo-V2.5-Pro costs less
Repository review
50K input + 3K output tokens
MiMo-V2.5-Pro costs less
Agent loop
200K input + 10K output tokens
MiMo-V2.5-Pro 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 | NVIDIA Nemotron 3 Ultra | Xiaomi MiMo-V2.5-Pro |
|---|---|---|
| Maker | NVIDIA | Xiaomi |
| Weights | Open weights | Open weights |
| Released | June 4, 2026 | April 22, 2026 |
| Context window | 262K | 1.05M |
| Input $/Mtok | $0.60 | $0.4575 |
| Output $/Mtok | $2.40 | $0.9152 |
| Coding index | 30.0 | 31.6 |
| Independent measurements behind it | 3 | 4 |
| Cost per solved task | Not computable | $0.03 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Nemotron 3 Ultra better than MiMo-V2.5-Pro for coding?
Nemotron 3 Ultra and MiMo-V2.5-Pro are level for coding, 30.0 against 31.6 on the Vibe Coding Index, so price and context decide. Pick Nemotron 3 Ultra for work measured by LiveCodeBench (Vals AI) or GPQA Diamond, where it leads. Pick MiMo-V2.5-Pro for output tokens 2.6x cheaper ($0.9152 vs $2.40 per million) and a larger context window (1.05M against 262K tokens).
Which is cheaper, Nemotron 3 Ultra or MiMo-V2.5-Pro?
Nemotron 3 Ultra costs $0.60 per million input tokens and $2.40 per million output tokens; MiMo-V2.5-Pro costs $0.4575 and $0.9152. Cost per solved task is known for MiMo-V2.5-Pro only.
Which has the bigger context window, Nemotron 3 Ultra or MiMo-V2.5-Pro?
MiMo-V2.5-Pro, with 1.05M tokens against 262K.
How many benchmarks is the Nemotron 3 Ultra vs MiMo-V2.5-Pro comparison based on?
62 benchmarks have a result for both models as of October 22, 2026. Nemotron 3 Ultra's coding index rests on 3 independent benchmarks and MiMo-V2.5-Pro's on 4. Where only one model has a result, the page says so instead of naming a winner.