Microsoft’s O3-Mini-High Model Now Available for Free in Copilot as AI Reasoning Capabilities Expand

Microsoft has significantly upgraded its Copilot service by providing users with free, unlimited access to the o3-mini-high model, enhancing the AI assistant’s reasoning capabilities. This development comes amid a wave of advancements in AI reasoning models across the industry.

Microsoft Enhances Copilot with Advanced Reasoning

Microsoft announced that its “Think Deeper” feature in Copilot is now powered by the o3-mini-high model, offering users improved reasoning capabilities without additional cost. This upgrade allows Copilot to engage in more extended chains of thought before providing answers, resulting in more accurate and nuanced responses.

“Think Deeper just got smarter. Now powered by o3-mini-high free in Copilot,” Microsoft executive Yusuf Mehdi shared on social media, highlighting the company’s commitment to making advanced AI accessible to all users.

The o3-mini-high model represents a significant step forward in Microsoft’s AI offerings, particularly in its ability to engage in deep reasoning processes. Users can now benefit from more thorough analysis and problem-solving capabilities without the premium price tag that often accompanies cutting-edge AI technologies.

The Rise of Reasoning Models

The integration of o3-mini-high into Copilot is part of a broader industry shift toward reasoning-focused AI models. These models differ from their predecessors by explicitly walking through step-by-step thought processes before reaching conclusions, similar to human reasoning.

DeepSeek has emerged as another significant player in this space, with its models demonstrating impressive efficiency improvements. The company’s approach has reportedly achieved a 57x efficiency boost through its mathematical language agent (MLA) technology, allowing for more powerful reasoning with fewer computational resources.

The industry consensus appears to be shifting toward viewing non-reasoning models as increasingly obsolete. As one source noted, “Non-reasoning models are obsolete and should disappear,” reflecting the growing importance of explicit reasoning capabilities in modern AI systems.

Competitive Landscape in AI Development

The AI assistant market continues to evolve rapidly, with multiple models now available to users:

– **Microsoft Copilot**: Now featuring o3-mini-high for enhanced reasoning
– **QwQ 32B**: Recently made available on Qwen chat
– **DeepSeek R1**: Known for its reasoning capabilities and competitive pricing
– **Claude 3.7 Sonnet**: Anthropic’s reasoning-capable model
– **GPT-4.5**: OpenAI’s latest model showing strong performance in certain benchmarks

Benchmark results across these models have shown varying strengths. While GPT-4.5 has demonstrated impressive results in coding and long-context comprehension tasks, reasoning-specific models like Claude-3.7-Sonnet-Thinking have shown superior performance in other areas.

One Reddit user who averaged performance across 11 different benchmarks found that “Claude-3.7-Sonnet-Thinking scored 69.41%, followed by GPT-4.5-Preview at 66.26%, and Claude-3.7-Sonnet at 61.63%,” illustrating the competitive nature of the current AI landscape.

Challenges in Evaluating Reasoning Models

Despite the excitement surrounding these advancements, experts have identified challenges in properly evaluating reasoning models. One significant issue is the varying token usage across different models to solve identical tasks.

“After seeing some initial thoughts on QwQ, it seems that the consensus is that it is a solid model, but utilizes 2-3x tokens to solve tasks vs R1. And then if we look at R1 vs o3-mini, R1 is the one that often utilizes more tokens to solve the same tasks,” noted one analyst.

This discrepancy affects not only pricing when deploying these models but also practical considerations like latency. Some experts have called for benchmarks that control for token usage to provide a more accurate picture of model efficiency and cost-effectiveness.

Future Directions

The rapid development of reasoning models suggests that AI capabilities will continue to expand in the near future. Researchers are exploring concepts like allowing AI to “reason infinitely” for complex problems, potentially enabling models to review and refine their thinking over extended periods.

Some are also investigating methods to dynamically control reasoning effort through system prompts, potentially allowing users to specify how deeply an AI should think about a given problem.

With Microsoft making advanced reasoning capabilities freely available through Copilot, we may see other providers follow suit, democratizing access to increasingly sophisticated AI tools.

As reasoning models continue to evolve, their applications across industries from healthcare to education are likely to expand, potentially transforming how we interact with AI assistants in our daily lives.