Microsoft model product
Phi-3.5-MoE-instruct is a mixture-of-experts model with ~42B total parameters (6.6B active) and a 128K context window. It excels at reasoning, math, coding, and multilingual tasks, outperforming larger dense models in many benchmarks. It underwent a thorough safety post-training process (SFT + DPO) and is licensed under MIT. This model is ideal for scenarios where efficiency and high performance are both required, particularly in multi-lingual or reasoning-intensive tasks.
Updated Aug 10, 2026. Default version: Phi-3.5-MoE-instruct
Structured fields from the published default version.
This profile uses the latest version under this unique model that has a calculated LLMBoard score. Arena and price are excluded.
Published benchmark records for the scored version Phi-3.5-MoE-instruct.
| GovReport | 0.3 | 1 | 2 | 100.0% | C | |
| MEGA MLQA | 0.7 | 1 | 2 | 100.0% | C | |
| MEGA TyDi QA | 0.7 | 1 | 2 | 100.0% | C | |
| MEGA UDPOS | 0.6 | 1 | 2 | 100.0% | C | |
| MEGA XCOPA | 0.8 | 1 | 2 | 100.0% | C | |
| MEGA XStoryCloze | 0.8 | 1 | 2 | 100.0% | C | |
| OpenBookQA | 0.9 | 1 | 5 | 100.0% | C | |
| PIQA | 0.9 | 1 | 11 | 100.0% | C | |
| RepoQA | 0.8 | 1 | 2 | 100.0% | C | |
| Social IQa | 0.8 | 1 | 9 | 100.0% | C | |
| SummScreenFD | 0.2 | 1 | 2 | 100.0% | C | |
| Qasper | 0.4 | 2 | 2 | 0.0% | C | |
| QMSum | 0.2 | 2 | 2 | 0.0% | C | |
| SQuALITY | 0.2 | 2 | 5 | 75.0% | C | |
| TruthfulQA | 0.8 | 2 | 18 | 94.1% | C | |
| BoolQ | 0.8 | 3 | 10 | 77.8% | C | |
| RULER | 0.9 | 3 | 4 | 33.3% | C | |
| BIG-Bench Hard | 0.8 | 9 | 21 | 60.0% | C | |
| Winogrande | 0.8 | 9 | 22 | 61.9% | C | |
| ARC-C | 0.9 | 10 | 34 | 72.7% | C | |
| MBPP | 0.8 | 12 | 37 | 69.4% | C | |
| HellaSwag | 0.8 | 14 | 27 | 50.0% | C | |
| Arena Hard | 0.4 | 22 | 26 | 16.0% | C | |
| MGSM | 0.6 | 26 | 31 | 16.7% | C | |
| GSM8k | 0.9 | 30 | 48 | 38.3% | C | |
| MMMLU | 0.7 | 44 | 49 | 10.4% | C | |
| MATH | 0.6 | 49 | 71 | 31.4% | C | |
| HumanEval | 0.7 | 64 | 76 | 16.0% | C | |
| MMLU | 0.8 | 69 | 100 | 31.3% | C | |
| MMLU-Pro | 0.5 | 120 | 129 | 7.0% | C | |
| GPQA | 0.4 | 210 | 233 | 9.9% | C |
Preference and agent-evaluation signals from published Arena datasets.
The default version has no published Arena rows, or its source alias has not been resolved.
Official vendor API PAYG pricing is summarized first. The table then lists individual provider offerings without treating their minimum as the official price.
The default version has no provider offering with current input or output token prices.
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
All published versions linked to this unique model. The score columns identify the version used by the current overall ranking.
| Phi-3.5-MoE-instruct | 42.4 | 60B | N/A | N/A | No | MIT |
A concise description based on the published model registry.
Phi-3.5-MoE-instruct is a mixture-of-experts model with ~42B total parameters (6.6B active) and a 128K context window. It excels at reasoning, math, coding, and multilingual tasks, outperforming larger dense models in many benchmarks. It underwent a thorough safety post-training process (SFT + DPO) and is licensed under MIT. This model is ideal for scenarios where efficiency and high performance are both required, particularly in multi-lingual or reasoning-intensive tasks.
Use the benchmark, Arena and pricing sections above as separate evidence. A missing field means the current data snapshot does not support that claim.
Data snapshot: 2026-08-07. Editorial model content is not available in the backend.
Open a comparison with the three ranked models immediately above and below this model.
Recommendations prioritize the same model type and family, then the closest published LLMBoard score.
Common questions about Phi 3.5 MoE.
Phi 3.5 MoE's default version was released on Aug 23, 2024.
No official standard PAYG price is currently available for Phi 3.5 MoE.
Phi 3.5 MoE is published under Microsoft in the model registry.
The current registry does not publish a context window for the default version.
No. The default version is not marked as having publicly available weights.
No published provider offering is currently linked to the default version.
Nearby ranked alternatives include Mistral Small 3.1 24B, GPT-5.4-nano, Qwen3 VL 235B A22B.