Microsoft Azure: Models Intelligence, Performance & Price
This analysis is intended to support you in choosing the best model provided by Microsoft Azure for your use-case.
Most Intelligent
Claude Opus 5.5 (max)
GPT-6 Astra (max)GPT-6.1 Sol (max)Claude Sonnet 5.5 (max)GPT-6 Sol (max)Intelligence Index
Total 78 models
Lowest Time per Task
GPT-6 Luna (non-reasoning)Claude Sonnet 5.5 (low)Claude Opus 5.5 (low)GPT-6 Astra (low)Claude 4.5 HaikuTime per Intelligence Index task (minutes)
Total 78 models
Lowest Price
GPT-5 nano (high)GPT-6 Luna (max)GPT-6 Luna (non-reasoning)GPT-4.1 nanoPhi-4 MiniBlended price (per 1M tokens)
Total 78 models
Azure offers 78 models, each with different intelligence, performance, and pricing characteristics. Below is a comparison of the key metrics across models.
- For intelligence, the top models on Azure are Claude Opus 5.5 (max) (59), GPT-6 Astra (max) (57), and GPT-6.1 Sol (max) (55).
- For time per task, GPT-6 Luna (non-reasoning) (1.8 min), Claude Sonnet 5.5 (low) (2.5 min), and Claude Opus 5.5 (low) (2.9 min) finish an Intelligence Index task fastest on Azure. Time per task varies significantly across models, with the slowest taking 2.1x longer than the quickest.
- For latency, Phi-4 Multimodal (0.43s), Claude 4.5 Haiku (non-reasoning) (1.11s), and GPT-5.4 mini (non-reasoning) (1.24s) offer the lowest time to first answer token.
- For pricing, GPT-5 nano (high) ($0.05), GPT-6 Luna (max) ($0.08), and GPT-6 Luna (non-reasoning) ($0.08) offer the lowest blended prices per 1M tokens.
- For context window size, GPT-6 Sol (max) (1M), GPT-5.4 (xhigh) (1M), and GPT-6.1 Sol (max) (1M) support the largest context windows on Azure.
Intelligence
Artificial Analysis Intelligence Index
Evaluations
Agentic knowledge work, Elo
Knowledge
1 - hallucination rate
Agentic SaaS workflows
Agentic scientific research workflows in a terminal
Agentic software engineering in open-source repositories
Reasoning & knowledge
Professional document reasoning, All-pass
Agentic coding & terminal use
Physics reasoning
Quantitative analysis on spreadsheets & documents
Legal agentic work, Hallucination-Gated All-Pass Rate
Agentic business operations
Agentic real-world work tasks, Elo
Agentic interactive 3D experiences, Elo
Product and brand videos designed in code, Elo
Kubernetes incident root-cause analysis
Medical long context reasoning
Visual reasoning
Rich, varied, well-crafted language
Well-structured, scannable, answer-first
Truthful, acknowledges uncertainty, non-sycophantic
Use of filler and generic language
Intelligence Index vs. Price
Pricing
Cost per Intelligence Index Task
Intelligence Index vs. Price
Speed
Average of the single-request and multi-turn agentic workloads measured
Time per Intelligence Index Task
Quality
Endpoint Accuracy Index
Latency
Average of the single-request and multi-turn agentic workloads measured
Time to First Answer Token
Reliability
Share of requests the endpoint failed, across all the single-request and multi-turn agentic workloads measured
Endpoint Error Rate
Caching
Cache Hit Rate
Cost per Task vs. Cache Hit Rate
Further Analysis | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
Claude Opus 5.5 (max) | 59 | $11.43 | 27.7 | 392.90 | 10.4% | 1M | Proprietary | |||
GPT-6 Astra (max) | 57 | $6.46 | 21.6 | 211.10 | 2.1% | 524k | Proprietary | |||
GPT-6.1 Sol (max) | 55 | $1.29 | 27.5 | 204.36 | 1.0% | 1M | Proprietary | |||
Claude Sonnet 5.5 (max) | 55 | $8.91 | 31.7 | 249.63 | 7.0% | 1M | Proprietary | |||
GPT-6 Sol (max) | 44 | $2.52 | 8.4 | 56.07 | 0.0% | 1M | Proprietary | |||
GPT-6 Astra (low) | 43 | $1.47 | 3.5 | 3.06 | 0.8% | 524k | Proprietary | |||
Claude Opus 5.5 (low) | 35 | $1.07 | 2.9 | 6.70 | 0.0% | 1M | Proprietary | |||
GPT-6 Luna (max) | 28 | $0.16 | 11.5 | 78.57 | 0.0% | 524k | Proprietary | |||
Claude Sonnet 5.5 (low) | 26 | $0.48 | 2.5 | 2.80 | 0.7% | 1M | Proprietary | |||
GPT-6 Luna (high) | 21 | $0.08 | -- | -- | -- | 524k | Proprietary | |||
GPT-6 Luna (non-reasoning) | 9 | $0.08 | 1.8 | 1.58 | 1.4% | 524k | Proprietary | |||
Claude 4.5 Haiku | 7 | $0.50 | 3.8 | 18.89 | 0.0% | 200k | Proprietary | |||
gpt-oss-120b (high) | 4 | $0.35 | 5.5 | 11.42 | 10.8% | 131k | Open | |||
GPT-5.2 Codex (xhigh) | -- | -- | -- | 72.22 | 0.0% | 400k | Proprietary | |||
GPT-5.1 Codex (high) | -- | -- | -- | 13.77 | 0.0% | 400k | Proprietary | |||
GPT-5 nano (minimal) East US 2 - Global Standard | -- | -- | -- | -- | -- | 400k | Proprietary | |||
o3-mini (high) | -- | -- | -- | 17.62 | 0.0% | 200k | Proprietary | |||
o1-preview | -- | -- | -- | -- | -- | 128k | Proprietary | |||
 Kimi K2.6 | -- | -- | -- | 18.54 | 0.0% | 262k | Open | |||
Kimi K2.6 (non-reasoning) | -- | -- | -- | 2.86 | 0.0% | 262k | Open | |||
o3 | -- | -- | -- | 17.72 | 0.0% | 200k | Proprietary | |||
Claude Sonnet 5 (max) | -- | -- | -- | 139.67 | 0.0% | 1M | Proprietary | |||
 Grok 3 | -- | -- | -- | -- | -- | 16k | Proprietary | |||
Claude Opus 4.5 | -- | -- | -- | 13.59 | 0.0% | 200k | Proprietary | |||
GPT-5 mini (high) | -- | -- | -- | 42.55 | 0.0% | 400k | Proprietary | |||
Phi-4 Multimodal | -- | -- | -- | 0.43 | 18.8% | 4k | Open | |||
o3-mini | -- | -- | -- | 6.28 | 0.0% | 200k | Proprietary | |||
Grok 4 Fast (non-reasoning) | -- | -- | -- | -- | -- | 2M | Proprietary | |||
Claude Sonnet 4.6 (non-reasoning, high) | -- | -- | -- | 1.56 | 0.0% | 200k | Proprietary | |||
GPT-5.4 mini (medium) | -- | -- | -- | -- | -- | 400k | Proprietary | |||
GPT-4.1 | -- | -- | -- | 1.38 | 0.0% | 1M | Proprietary | |||
 Command A | -- | -- | -- | 2.22 | 8.6% | 256k | Open | |||
Claude Opus 4.5 (non-reasoning) | -- | -- | -- | 1.61 | 0.0% | 200k | Proprietary | |||
Claude Opus 4.6 (non-reasoning, high) | -- | -- | -- | 2.08 | 2.6% | 1M | Proprietary | |||
GPT-5 (high) | -- | -- | -- | 38.03 | 0.0% | 400k | Proprietary | |||
 Mistral Large 3 | -- | -- | -- | 3.69 | 0.0% | 256k | Open | |||
GPT-5.2 (xhigh) | -- | -- | -- | 77.10 | 0.0% | 400k | Proprietary | |||
GPT-5.1 (high) | -- | -- | -- | 45.28 | 2.4% | 400k | Proprietary | |||
Claude Opus 4.6 (max) | -- | -- | -- | 39.19 | 0.0% | 1M | Proprietary | |||
Grok 4 Fast | -- | -- | -- | -- | -- | 2M | Proprietary | |||
Mistral Medium 3 | -- | -- | -- | 5.71 | 2.6% | 128k | Proprietary | |||
o1 | -- | -- | -- | -- | -- | 200k | Proprietary | |||
GPT-5 Codex (high) | -- | -- | -- | 10.30 | 0.0% | 400k | Proprietary | |||
 DeepSeek V4 Pro (high) | -- | -- | -- | -- | -- | 1M | Open | |||
Grok 4.3 (medium) | -- | -- | -- | -- | -- | 200k | Proprietary | |||
GPT-5 (medium) | -- | -- | -- | -- | -- | 400k | Proprietary | |||
Grok 4 | -- | -- | -- | 18.60 | 0.0% | 256k | Proprietary | |||
DeepSeek V3.2 (non-reasoning) | -- | -- | -- | 1.41 | 0.0% | 128k | Open | |||
GPT-5.2 (non-reasoning) | -- | -- | -- | 2.07 | 0.0% | 400k | Proprietary | |||
DeepSeek V4 Pro (non-reasoning) | -- | -- | -- | 3.97 | 0.0% | 1M | Open | |||
GPT-4.1 nano | -- | -- | -- | 1.61 | 0.0% | 1M | Proprietary | |||
GPT-4 Turbo | -- | -- | -- | 1.66 | 0.0% | 128k | Proprietary | |||
GPT-5 (low) | -- | -- | -- | -- | -- | 400k | Proprietary | |||
GPT-5.6 Terra (high) | -- | -- | -- | 4.76 | 0.0% | 524k | Proprietary | |||
o4-mini (high) | -- | -- | -- | 21.02 | 0.0% | 200k | Proprietary | |||
GPT-5.2 (medium) | -- | -- | -- | -- | -- | 400k | Proprietary | |||
GPT-5.6 Luna (xhigh) | -- | -- | -- | 25.66 | 0.0% | 524k | Proprietary | |||
GPT-5.6 Sol (high) | -- | -- | -- | -- | -- | 524k | Proprietary | |||
GPT-4o (Aug) | -- | -- | -- | 2.12 | 0.0% | 128k | Proprietary | |||
GPT-5 nano (medium) | -- | -- | -- | -- | -- | 400k | Proprietary | |||
Claude 4.5 Sonnet | -- | -- | -- | 17.37 | 0.0% | 200k | Proprietary | |||
Claude 4.5 Sonnet (non-reasoning) | -- | -- | -- | 1.83 | 0.0% | 200k | Proprietary | |||
 Llama 4 Maverick (FP8) | -- | -- | -- | 1.91 | 11.7% | 128k | Open | |||
Grok 4.3 (high) | -- | -- | -- | 8.31 | 0.0% | 200k | Proprietary | |||
Llama 3.3 70B | -- | -- | -- | 1.85 | 3.6% | 128k | Open | |||
GPT-5.6 Terra (xhigh) | -- | -- | -- | 16.72 | 0.0% | 524k | Proprietary | |||
GPT-4.1 mini | -- | -- | -- | 1.45 | 0.0% | 1M | Proprietary | |||
Phi-4 Mini | -- | -- | -- | 4.61 | 28.2% | 128k | Open | |||
Kimi K2.5 | -- | -- | -- | 14.60 | 0.0% | 262k | Open | |||
GPT-5.4 (xhigh) | -- | -- | -- | 73.65 | 0.0% | 1M | Proprietary | |||
GPT-5.6 Luna (high) | -- | -- | -- | 8.21 | 0.0% | 524k | Proprietary | |||
Llama 4 Scout | -- | -- | -- | 1.55 | 0.0% | 128k | Open | |||
GPT-5.4 mini (non-reasoning) | -- | -- | -- | 1.24 | 0.0% | 400k | Proprietary | |||
GPT-5.5 (high) | -- | -- | -- | 16.83 | 0.0% | 524k | Proprietary | |||
GPT-4o mini | -- | -- | -- | 2.17 | 0.0% | 128k | Proprietary | |||
Kimi K2.5 (non-reasoning) | -- | -- | -- | 3.66 | 0.0% | 262k | Open | |||
GPT-5.4 mini (xhigh) | -- | -- | -- | 103.06 | 0.0% | 400k | Proprietary | |||
GPT-5 mini (minimal) East US 2 - Global Standard | -- | -- | -- | -- | -- | 400k | Proprietary | |||
Kimi K2 Thinking | -- | -- | -- | 13.86 | 0.0% | 256k | Open | |||
DeepSeek V4 Pro (max) | -- | -- | -- | 16.81 | 0.0% | 1M | Open | |||
GPT-4o (Nov) | -- | -- | -- | 1.48 | 0.0% | 128k | Proprietary | |||
Claude 4.5 Haiku (non-reasoning) | -- | -- | -- | 1.11 | 0.0% | 200k | Proprietary | |||
GPT-5 mini (medium) | -- | -- | -- | -- | -- | 400k | Proprietary | |||
GPT-5 (minimal) | -- | -- | -- | 2.54 | 0.0% | 400k | Proprietary | |||
Grok 4.20 0309 v2 | -- | -- | -- | 7.79 | 5.1% | 262k | Proprietary | |||
gpt-oss-120b (low) | -- | -- | -- | 2.08 | 9.4% | 131k | Open | |||
o3-pro | -- | -- | -- | -- | -- | 200k | Proprietary | |||
Claude 4.1 Opus | -- | -- | -- | -- | -- | 200k | Proprietary | |||
Claude Fable 5 | -- | -- | -- | 93.09 | 0.0% | 1M | Proprietary | |||
GPT-5.1 (non-reasoning) | -- | -- | -- | 2.08 | 0.0% | 400k | Proprietary | |||
GPT-5.6 Sol (max) | -- | -- | -- | 62.31 | 0.0% | 524k | Proprietary | |||
GPT-5.6 Sol (xhigh) | -- | -- | -- | -- | -- | 524k | Proprietary | |||
Claude Sonnet 4.6 (max) | -- | -- | -- | 107.93 | 0.0% | 200k | Proprietary | |||
GPT-5 nano (high) | -- | -- | -- | 39.65 | 0.0% | 400k | Proprietary | |||
GPT-4o (May) | -- | -- | -- | 2.10 | 0.0% | 128k | Proprietary | |||
Claude Opus 4.7 (max) | -- | -- | -- | 18.11 | 0.0% | 1M | Proprietary | |||
Phi-4 | -- | -- | -- | 2.33 | 0.0% | 16k | Open | |||
Claude 4.1 Opus (non-reasoning) | -- | -- | -- | -- | -- | 200k | Proprietary | |||
GPT-5.1 Codex mini (high) | -- | -- | -- | 18.00 | 0.0% | 400k | Proprietary | |||
Grok 4.3 (low) | -- | -- | -- | -- | -- | 200k | Proprietary | |||
Grok 3 mini Reasoning (high) | -- | -- | -- | -- | -- | 32k | Proprietary | |||
Key definitions
Frequently Asked Questions
Common questions about Microsoft Azure
Microsoft Azure offers 78 models that we track: Claude Opus 5.5 (max), GPT-6 Astra (max), GPT-6.1 Sol (max), Claude Sonnet 5.5 (max), GPT-6 Sol (max), GPT-6 Astra (low), Claude Opus 5.5 (low), GPT-6 Luna (max), Claude Sonnet 5.5 (low), GPT-6 Luna (non-reasoning), Claude 4.5 Haiku, gpt-oss-120b (high), GPT-5.2 Codex (xhigh), GPT-5.1 Codex (high), o3-mini (high), Kimi K2.6, Kimi K2.6 (non-reasoning), o3, Claude Sonnet 5 (max), Claude Opus 4.5, GPT-5 mini (high), Phi-4 Multimodal, o3-mini, Claude Sonnet 4.6 (non-reasoning, high), GPT-4.1, Command A, Claude Opus 4.5 (non-reasoning), Claude Opus 4.6 (non-reasoning, high), GPT-5 (high), Mistral Large 3, GPT-5.2 (xhigh), GPT-5.1 (high), Claude Opus 4.6 (max), Mistral Medium 3, GPT-5 Codex (high), Grok 4, DeepSeek V3.2 (non-reasoning), GPT-5.2 (non-reasoning), DeepSeek V4 Pro (non-reasoning), GPT-4.1 nano, GPT-4 Turbo, GPT-5.6 Terra (high), o4-mini (high), GPT-5.6 Luna (xhigh), GPT-4o (Aug), Claude 4.5 Sonnet, Claude 4.5 Sonnet (non-reasoning), Llama 4 Maverick (FP8), Grok 4.3 (high), Llama 3.3 70B, GPT-5.6 Terra (xhigh), GPT-4.1 mini, Phi-4 Mini, Kimi K2.5, GPT-5.4 (xhigh), GPT-5.6 Luna (high), Llama 4 Scout, GPT-5.4 mini (non-reasoning), GPT-5.5 (high), GPT-4o mini, Kimi K2.5 (non-reasoning), GPT-5.4 mini (xhigh), Kimi K2 Thinking, DeepSeek V4 Pro (max), GPT-4o (Nov), Claude 4.5 Haiku (non-reasoning), GPT-5 (minimal), Grok 4.20 0309 v2, gpt-oss-120b (low), Claude Fable 5, GPT-5.1 (non-reasoning), GPT-5.6 Sol (max), Claude Sonnet 4.6 (max), GPT-5 nano (high), GPT-4o (May), Claude Opus 4.7 (max), Phi-4, and GPT-5.1 Codex mini (high).
The most intelligent model available on Microsoft Azure is Claude Opus 5.5 (max) with an Intelligence Index score of 59.
The model with the lowest time per task on Microsoft Azure is GPT-6 Luna (non-reasoning) at 1.8 minutes per Intelligence Index task. Time per task is the output tokens per task divided by end-to-end output speed, so lower means a task finishes sooner.
The model with the lowest time to first answer token on Microsoft Azure is Phi-4 Multimodal at 0.43s. Lower latency means faster initial response time.
The most affordable model on Microsoft Azure by blended price is GPT-5 nano (high) at $0.0535 per 1M tokens (7:2:1 cache hit/input/output ratio).
Prices on Microsoft Azure vary up to 224.3x across models, from $0.0535 per 1M tokens for GPT-5 nano (high) to $12.00 per 1M tokens for GPT-4 Turbo.
Yes, Microsoft Azure offers an OpenAI-compatible API, making it easy to switch from OpenAI or use existing OpenAI SDK integrations.
61 of 78 models on Microsoft Azure support JSON mode for structured output.
75 of 78 models on Microsoft Azure support function calling (tool use).
Yes, Microsoft Azure offers 47 reasoning models: Claude Opus 5.5 (max), GPT-6 Astra (max), GPT-6.1 Sol (max), Claude Sonnet 5.5 (max), GPT-6 Sol (max), GPT-6 Astra (low), Claude Opus 5.5 (low), GPT-6 Luna (max), Claude Sonnet 5.5 (low), Claude 4.5 Haiku, gpt-oss-120b (high), GPT-5.2 Codex (xhigh), GPT-5.1 Codex (high), o3-mini (high), Kimi K2.6, o3, Claude Sonnet 5 (max), Claude Opus 4.5, GPT-5 mini (high), o3-mini, GPT-5 (high), GPT-5.2 (xhigh), GPT-5.1 (high), Claude Opus 4.6 (max), GPT-5 Codex (high), Grok 4, GPT-5.6 Terra (high), o4-mini (high), GPT-5.6 Luna (xhigh), Claude 4.5 Sonnet, Grok 4.3 (high), GPT-5.6 Terra (xhigh), Kimi K2.5, GPT-5.4 (xhigh), GPT-5.6 Luna (high), GPT-5.5 (high), GPT-5.4 mini (xhigh), Kimi K2 Thinking, DeepSeek V4 Pro (max), Grok 4.20 0309 v2, gpt-oss-120b (low), Claude Fable 5, GPT-5.6 Sol (max), Claude Sonnet 4.6 (max), GPT-5 nano (high), Claude Opus 4.7 (max), and GPT-5.1 Codex mini (high). Reasoning models use extended thinking to work through complex problems before providing an answer.
Yes, 18 of 78 models on Microsoft Azure are open weight models: gpt-oss-120b (high), Kimi K2.6, Kimi K2.6 (non-reasoning), Phi-4 Multimodal, Command A, Mistral Large 3, DeepSeek V3.2 (non-reasoning), DeepSeek V4 Pro (non-reasoning), Llama 4 Maverick (FP8), Llama 3.3 70B, Phi-4 Mini, Kimi K2.5, Llama 4 Scout, Kimi K2.5 (non-reasoning), Kimi K2 Thinking, DeepSeek V4 Pro (max), gpt-oss-120b (low), and Phi-4.
Yes, provider performance can vary over time due to infrastructure changes, load balancing, and updates. We continuously benchmark all providers and display historical performance trends in the "Over Time" charts.
When choosing a model on Microsoft Azure, consider: intelligence (for quality-sensitive tasks), time per task (how quickly the model completes an Intelligence Index task), latency (for interactive applications requiring quick first responses), pricing (for cost-sensitive workloads), and features like context window size, JSON mode, or function calling support.