Executive Summary

This week saw significant developments in AI models and legal industry adoption, including a major law firm AI integration and new releases from Google and OpenAI.

Key Developments

1. Freshfields Partners with Google Cloud on AI Integration

Global law firm Freshfields Bruckhaus Deringer announced a strategic collaboration with Google Cloud to accelerate AI adoption in its legal services. This partnership represents one of the first major law firms publicly deploying cutting-edge AI at scale directly from a major AI model provider. The firm will integrate Gemini AI across its workflows and develop custom AI agents for legal processes, including a "Dynamic Due Diligence" tool for contract review.^[1]

2. AI Model Rankings

Google's Gemini 2.5 Pro currently holds the #1 position on the LLM Arena leaderboard with a significant margin. OpenAI's ChatGPT-4o jumped from #5 to #2 in recent rankings. For factual accuracy, Google's Gemini models have shown the strongest performance on benchmarks focused on grounding responses in provided source material, performing well on both the FACTS Grounding Leaderboard and the Vectara Hallucination Evaluation Leaderboard.^[2]

3. Google Launches Gemini 2.5 Flash with Adjustable Reasoning

Google unveiled Gemini 2.5 Flash, featuring a "thinking budget" that allows users to control reasoning depth. This provides a balance between response quality, speed, and computational cost.^[3]

4. OpenAI Releases o4-mini Lightweight Reasoning Model

OpenAI launched two new reasoning models: flagship o3 and smaller o4-mini. These models can autonomously leverage all ChatGPT tools (web browsing, code execution, image analysis) to solve multi-step problems.^[4]

5. ChatGPT-4o's Praise Bias Raises Concerns

Users have reported that OpenAI's ChatGPT-4o model now tends to respond with excessive positivity, often over-complimenting user input regardless of quality.^[5] Why It Matters for Lawyers: When using AI for drafting or document review, this "praise bias" could mask critical flaws or issues that require attention. Legal professionals seeking objective critique should specifically prompt for neutrality or consider alternative models like Gemini, Claude, or Grok for more direct feedback on critical written work.^[5]

Conclusion

The rapid evolution of AI continues to present new opportunities for legal professionals, with this week's developments highlighting both the advancement of AI capabilities and the growing adoption of these tools in mainstream legal settings.

Conflicts of Interest. The author practiced law for six years at Ropes & Gray LLP prior to co-founding UseJunior. UseJunior is a member of both Google for Startups and Microsoft for Startups and has received grants and in-kind contributions in connection with these programs. The views expressed herein are our own and do not represent the views of Google, Microsoft or Ropes & Gray LLP. This is not legal advice.

References

1. Freshfields, Freshfields and Google Cloud Accelerate Legal Innovation Through Strategic AI Collaboration, FRESHFIELDS (Apr. 8, 2025), https://www.freshfields.com/en/our-thinking/news/news-search/2025/04/freshfields-and-google-cloud-accelerate-legal-innovation-through-strategic-ai-collaboration/. Freshfields reports it will integrate Google's Gemini AI via Google Workspace throughout its internal workflows, embedding generative AI into tools like Google Docs, Sheets, and NotebookLM. Additionally, the firm plans to leverage Google Cloud's Vertex AI platform to develop bespoke AI agents and solutions, including the "Dynamic Due Diligence" tool powered by Gemini's advanced reasoning capabilities combined with secure search and firm data using Google's Agentspace framework.

2. Chatbot Arena, LM ARENA, https://lmarena.ai/ (last visited Apr. 18, 2025); The FACTS Grounding Leaderboard, KAGGLE, https://www.kaggle.com/facts-leaderboard (last visited Apr. 18, 2025). (See also Chatbot Arena Leaderboard, HUGGING FACE, https://huggingface.co/spaces/lmarena-ai/chatbot-arena-leaderboard; Hughes Hallucination Evaluation Model (HHEM) leaderboard, HUGGING FACE, HHEM Leaderboard - a Hugging Face Space by vectara.)

3. Start Building with Gemini 2.5 Flash, GOOGLE DEVS. BLOG (Apr. 17, 2025), https://developers.googleblog.com/en/start-building-with-gemini-25-flash/. Google's announcement highlighted how the "thinking budget" feature represents a significant advancement in controllable AI reasoning. The feature allows developers to specify how much computation the model should dedicate to reasoning through complex problems.

4. OpenAI, Introducing OpenAI o3 and o4-mini, OPENAI BLOG (Apr. 16, 2025), https://openai.com/index/introducing-o3-and-o4-mini/. OpenAI describes o4-mini as offering a competitive balance of speed, cost, and performance. Both models became available to ChatGPT Plus users around the announcement date. The autonomous tool use capability marks a significant advancement in multi-modal problem solving.

5. Nearcyan (@nearcyan), X (Apr. 17, 2025, 4:20 PM), https://x.com/nearcyan/status/1912952857571336299 (reporting ChatGPT-4o praise bias observation). As of mid-April 2025, OpenAI had not officially commented on this specific behavior. At UseJunior, there's a general preference for models like Gemini, Claude, and Grok when seeking direct and unvarnished criticism of critical written work product.