vs
0:11
overall capability wins (0 ties across 11 scored). MiMo-V2.5-Pro leads overall.
- Reasoning0:6
- Coding0:3
- Tool use0:1
- Composite indices0:2
- Specs0:1
Verdict
MiMo-V2.5-Pro leads coding (widest gap: +308.3 on WebDev Elo); MiMo-V2.5-Pro offers ~7.8x larger context (1M vs 128K); MiMo-V2.5-Pro ships open weights (MIT).
- MiMo-V2.5-Pro leads coding (widest gap: +308.3 on WebDev Elo)
- MiMo-V2.5-Pro offers ~7.8x larger context (1M vs 128K)
- MiMo-V2.5-Pro ships open weights (MIT)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Mercury 2
- $0.44
- MiMo-V2.5-Pro
- -
Price per 1M tokens
Full comparison
Organization
Inception Labs
Xiaomi
Family
Other
Other
License
Proprietary
MIT
Open weights
No
Yes
Release
Sep 1, 2026
Apr 27, 2026
Knowledge cutoff
-
-
API / provider
Inception Labs
No primary API price listed
Modalities
text → text
text → text
Specs
Context window
128K
1M
Max output
-
-
Parameters
—
1.0T (42B act.)
Pricing
Input $/1M
$0.25
—
Output $/1M
$0.75
—
Blended $/1M (3∶1)
$0.38
-
Speed
tok/s
-
-
TTFT (s)
-
-
Reasoning
MMLU-Pro
—
—
GPQA Diamond
77.0
86.6
Humanity's Last Exam
17.1
35.7
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
21.2
22.4
AA-LCR v1.1
43.7
79.7
CritPt
0.8
4.0
MMMU-Pro
—
—
IFBench
69.8
79.9
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
27.3
65.2
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
37.7
50.6
CursorBench
—
—
SWE-Rebench
—
44.4
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
1,167
1,475
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,467
| Metric | Mercury 2 | MiMo-V2.5-Pro | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Inception Labs | Xiaomi | — |
| Family | Other | Other | — |
| License | Proprietary | MIT | — |
| Open weights | No | Yes | — |
| Release | Sep 1, 2026 | Apr 27, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Inception Labs | No primary API price listed | — |
| Modalities | text → text | text → text | — |
| Specs | |||
| Context window | 128K | 1M | B +872K |
| Max output | - | - | — |
| Parameters | — | 1.0T (42B act.) | — |
| Pricing | |||
| Input $/1M | $0.25 | — | — |
| Output $/1M | $0.75 | — | — |
| Blended $/1M (3∶1) | $0.38 | - | — |
| Speed | |||
| tok/s | - | - | — |
| TTFT (s) | - | - | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 77.0 | 86.6 | B +9.6 pts |
| Humanity's Last Exam | 17.1 | 35.7 | B +18.6 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 21.2 | 22.4 | B +1.2 pts |
| AA-LCR v1.1 | 43.7 | 79.7 | B +36.0 pts |
| CritPt | 0.8 | 4.0 | B +3.2 pts |
| MMMU-Pro | — | — | — |
| IFBench | 69.8 | 79.9 | B +10.1 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 27.3 | 65.2 | B +37.9 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 37.7 | 50.6 | B +12.9 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | 44.4 | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | 1,167 | 1,475 | B +308 Elo |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,467 | — |
Mercury 2
- Context
- 128K
- Parameters
- —
- Price
- $0.25 / $0.75
- Speed
- —
- Modalities
- text → text
- License
- Proprietary
Inception Labs' diffusion-based (non-autoregressive) reasoning LLM, API-only with a 128K context window and high-throughput inference on Blackwell GPUs.
MiMo-V2.5-Pro
- Context
- 1M
- Parameters
- 1.0T (42B act.)
- Price
- —
- Speed
- —
- Modalities
- text → text
- License
- MIT
Xiaomi's flagship open-weight MoE model (1.02T total / 42B active parameters) in the MiMo-V2.5 series, tuned for coding, reasoning, and long-horizon agent tasks with a 1M-token context window, released under the MIT license.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.