Grok, Sonar, Copilot, and Phi — specs and best use cases for the remaining major providers.
Beyond Anthropic, OpenAI, Google, and Meta, several other providers offer notable models and distinctive value propositions. This article covers xAI (Grok), Perplexity, and Microsoft, each of which occupies a meaningful niche in the AI landscape.
---
xAI was founded by Elon Musk in 2023 after his departure from the OpenAI board. The company's models are branded Grok and are distinctive for their real-time access to X (formerly Twitter) data and a less restrictive content policy than most competitors.
| Property | Value | |---|---| | Context Window | 131,072 tokens (~98,000 words) | | Input Pricing (API) | $2.00 per million tokens | | Output Pricing (API) | $10.00 per million tokens | | Vision | Yes | | Tool Use | Yes | | Real-time X Data | Yes |
Best for: Applications that benefit from real-time social media data and current events. Users who want less content filtering than mainstream models.
Strengths: Access to real-time posts and trends from X (Twitter) is a genuine differentiator for current events, social sentiment analysis, and trend tracking. Vision capable. Competitive pricing.
Weaknesses: X data access creates its own noise — social media is not a reliable factual source. The X Premium+ consumer access requirement limits accessibility for non-subscribers. Less proven in enterprise deployments compared to established providers.
---
| Property | Value | |---|---| | Context Window | 131,072 tokens | | Input Pricing (API) | $3.00 per million tokens | | Output Pricing (API) | $15.00 per million tokens | | Vision | Yes | | Tool Use | Yes | | Real-time X Data | Yes | | Deep Search | Yes |
Best for: xAI's flagship model. Complex reasoning, research tasks, applications needing both strong intelligence and real-time data access.
Strengths: Significantly improved reasoning over Grok 2. xAI claims strong performance on hard reasoning benchmarks. "Deep Search" mode conducts extended web research before responding. Thinking mode for extended chain-of-thought reasoning.
Weaknesses: Context window is smaller than Anthropic or Google flagship models. xAI has less of a track record in enterprise deployments. Third-party benchmark validation is more limited than for more established models.
Access: X Premium+ subscription for consumer access via x.com/Grok. API available at console.x.ai.
---
Perplexity AI is not primarily a model company — it is a search and AI assistant product that wraps multiple underlying models (including its own Sonar models built on open-source foundations) around a real-time web retrieval system.
| Property | Value | |---|---| | Input Pricing | $1.00 per million tokens | | Output Pricing | $1.00 per million tokens | | Search Queries | $5.00 per 1,000 queries | | Citations | Yes — default | | Real-time Web Access | Yes — every response |
Best for: Factual question answering, research, current events. Any application where every answer should be grounded in retrievable sources.
Strengths: Every response cites sources by default — this is the entire product philosophy. Real-time web access means knowledge is always current. The citation system makes hallucinations detectable rather than invisible. Very different use model from pure language models.
Weaknesses: Not the right tool for creative writing, code generation, analysis, or tasks that do not benefit from web retrieval. Dependent on web search quality.
---
| Property | Value | |---|---| | Input Pricing | $3.00 per million tokens | | Output Pricing | $15.00 per million tokens | | Search Queries | $5.00 per 1,000 queries | | Citations | Yes — deeper research |
Best for: More complex research tasks requiring multiple search rounds, deeper synthesis, and higher-quality responses than Sonar standard.
Strengths: More search queries per response, better synthesis of multiple sources, stronger underlying model. Suitable for producing research-quality outputs with full citation trails.
Consumer Access: perplexity.ai — free tier available, Perplexity Pro at $20/month for higher usage, access to more models, and additional features including file upload and image generation.
---
Microsoft has invested approximately $13 billion in OpenAI and has integrated GPT-4o-class models into its products under the Copilot brand. Microsoft also develops its own Phi family of small language models for edge and on-device use.
| Property | Value | |---|---| | Underlying Model | GPT-4o (via OpenAI partnership) | | Access | copilot.microsoft.com, Windows 11, Edge browser | | Free Tier | Yes | | Copilot Pro | $20/month |
Best for: General AI assistance deeply integrated into Microsoft's product ecosystem. Users already in Microsoft 365 environment.
Strengths: Deeply integrated into Windows, Edge, Bing, and Microsoft 365 apps. Free tier with GPT-4o access is competitive. Image generation via DALL-E 3 included. Copilot+ PCs include on-device AI features.
Weaknesses: Less differentiated from OpenAI's own products than competitors with proprietary models. Product experiences have been inconsistent across different Microsoft surfaces.
---
| Property | Value | |---|---| | Pricing | $30/user/month (added to existing M365 subscription) | | Integration | Word, Excel, PowerPoint, Outlook, Teams, OneNote |
Best for: Enterprise productivity. Drafting emails in Outlook, summarizing meetings in Teams, generating documents in Word, analyzing data in Excel.
Strengths: Deep integration with enterprise data (SharePoint, emails, calendar, documents). Data stays within the organization's Microsoft tenant — important for compliance. Admin controls and data governance familiar to IT teams.
Weaknesses: Expensive on top of existing M365 licensing. Quality of suggestions varies significantly by task. Requires proper setup of information architecture to work well.
---
| Property | Value | |---|---| | Parameters | 14 billion | | Context Window | 16,384 tokens | | Input Pricing (Azure) | $0.07 per million tokens | | Output Pricing (Azure) | $0.14 per million tokens | | Open Weights | Yes (available on Hugging Face) |
Best for: On-device deployment, edge computing, applications requiring extreme cost efficiency, use cases where a 14B model with strong reasoning is sufficient.
Strengths: Microsoft's Phi series focuses on data quality over data quantity — training smaller models on carefully curated, high-quality data rather than maximizing data volume. Phi-4 punches above its weight class on reasoning benchmarks relative to parameter count. Very cheap via API. Runs on consumer hardware.
Weaknesses: 16K context window is small compared to other current models. Not suitable for long-document tasks. Smaller models have meaningful capability ceilings.
---
| Provider | Distinctive Advantage | Best Use Case | |---|---|---| | xAI (Grok) | Real-time X/social data, less restrictive | Current events, social intelligence | | Perplexity | Cited sources by default, search-grounded | Research, factual Q&A | | Microsoft Copilot | Office/Windows integration | Enterprise productivity in M365 | | Phi-4 | Efficient small model, on-device capable | Edge deployment, cost-sensitive |
Each of these providers occupies a real niche. Perplexity is legitimately different in its architecture (retrieval-first) rather than just positioning. xAI's real-time social data is a real capability gap. Microsoft's value is distribution and integration rather than model innovation. Phi is a genuinely interesting approach to efficiency.
Pricing pages: console.x.ai (xAI), docs.perplexity.ai/docs/pricing (Perplexity), azure.microsoft.com/pricing/details/cognitive-services/openai-service (Microsoft Azure)
Have a follow-up question about this topic?
Ask AI