microsoft/phi-3.5-moe-instruct · Microsoft Azure AI
16x3.8B MoE architecture (6.6B active) with 128K context window for fast edge server inference at $0.15/$0.60.
Last checked Aug 28, 2026. Calculations use the listed base rates; provider-specific tiers, cache writes, batch pricing, and long-context rules may differ. Confirm the current rate at the official provider source.
Quick answer
At the published base rate, Microsoft Phi-3.5 MoE costs $0.15 per 1M input tokens and $0.60 per 1M output tokens. It supports a 128K context window and is currently marked ga.
Method & trust
This rate card uses the provider's listed base input, output, and cached-input prices. Workload tables below apply those rates to explicit token counts, cache assumptions, and request volumes; verify provider-specific tiers before committing budget.
Input
$0.15
per 1M tokens
Output
$0.60
per 1M tokens
Cached input
—
not published
Context window
128K
max output 4.1K
Microsoft Phi-3.5 MoE cost calculator
What Microsoft Phi-3.5 MoE costs per task
| Use case | Input tokens | Output tokens | Cost / request | Monthly @ 1K req/day |
|---|---|---|---|---|
| Customer support chatbot | 3,500 | 350 | $0.000735 | $22.05 |
| RAG / search-augmented answers | 8,000 | 500 | $0.0015 | $45.00 |
| AI coding assistant | 12,000 | 2,000 | $0.003 | $90.00 |
| Document summarization | 25,000 | 600 | $0.00411 | $123.30 |
| Agentic workflow | 40,000 | 1,500 | $0.0069 | $207.00 |
| Content generation | 800 | 1,200 | $0.00084 | $25.20 |
| Data extraction & tagging | 2,000 | 250 | $0.00045 | $13.50 |
| Translation | 5,000 | 5,500 | $0.00405 | $121.50 |
Assumes each use case's typical cacheable share of input. See full cost scenarios.
Compare with alternatives
| Model | Input /M | Output /M | Context | Chat request* |
|---|---|---|---|---|
| Microsoft Phi-3.5 MoE | $0.15 | $0.60 | 128K | $0.00108 |
| GPT-5.6 Solcompare | $5.00 | $30.00 | 1.1M | $0.035 |
| GPT-5.6 Terracompare | $2.00 | $12.00 | 1.1M | $0.014 |
| GPT-5.6 Lunacompare | $0.20 | $1.20 | 1.1M | $0.0014 |
| GPT-5.6 Cybercompare | $12.50 | $75.00 | 1.1M | $0.0875 |
| GPT-5.5 Standardcompare | $5.00 | $30.00 | 512K | $0.035 |
| GPT-5.5 Procompare | $30.00 | $180.00 | 512K | $0.21 |
| GPT-5.4 Workhorsecompare | $2.50 | $15.00 | 256K | $0.0175 |
| GPT-5.4 minicompare | $0.75 | $4.50 | 256K | $0.00525 |
*4,000 in + 800 out tokens, 50% cached input where available.
Related calculations
FAQ
Microsoft Phi-3.5 MoE costs $0.15 per 1M input tokens and $0.60 per 1M output tokens. Verified Aug 28, 2026 against https://azure.microsoft.com/en-us/pricing/details/cognitive-services/.
Microsoft Phi-3.5 MoE supports a 128,000-token context window with up to 4,096 output tokens per request, tokenized with cl100k_base.
At $0.15/M input and $0.60/M output, Microsoft Phi-3.5 MoE sits below GPT-5.6 Sol ($5.00/M in, $30.00/M out) — see the comparison table for full-workload differences.