Perplexity Launches Web Browser ‘Comet’ as Anthropic’s Claude 3.7 Sonnet Gains Momentum

Tech Highlights for Today:

• Claude 3.7 Sonnet introduces ‘Thinking Mode’ for better AI reasoning.

• Perplexity unveils ‘Comet,’ an AI-powered web browser.

• Luma AI adds free AI-generated audio to DreamMachine video tool.

• OpenAI expands Deep Research access to Plus users.

• DeepSeek accelerates the launch of its next-gen AI model, R2.

Atla Unveils Selene 1, an Advanced LLM Judge for AI Evaluation

Atla has officially launched Selene 1, a cutting-edge large language model (LLM) designed to evaluate AI-generated responses for accuracy, quality, and coherence. Unlike traditional AI models that generate answers, Selene critiques and scores AI responses with superior efficiency.

Selene 1 has outperformed leading AI models, including OpenAI’s O-series, Anthropic’s Claude 3.5 Sonnet, and DeepSeek R1, across 11 industry-standard benchmarks for AI evaluation.

Key Features of Selene 1:

• Detects hallucinations and assesses correctness across diverse domains.

• Custom evaluation criteria enable developers to fine-tune assessments.

• Available via API/SDK for seamless integration.

• Supports Atla’s new Alignment Platform to refine AI evaluation metrics.

Developers can now leverage Selene 1 to ensure more reliable and accurate AI-generated content.

Claude 3.7 Sonnet Introduces ‘Thinking Mode’ for Adaptive AI Responses

Anthropic is redefining enterprise AI with Claude 3.7 Sonnet, an advanced model featuring ‘Thinking Mode’ — a toggle that allows users to switch between instant responses and deep analytical reasoning on demand.

Why It Matters:

• Benchmarks show Claude 3.7 achieves 78.2% accuracy on graduate-level reasoning tasks, surpassing DeepSeek R1 in instruction-following.

• Unlike competitors focusing on speed or cost-efficiency, Anthropic is emphasizing adaptive AI reasoning to balance depth and efficiency.

• The launch of Claude Code, an AI-powered command-line assistant, enhances AI oversight for developers.

By prioritizing context-aware intelligence over fragmented AI stacks, Anthropic aims to reshape how enterprises interact with AI.

Perplexity Unveils AI-Powered Web Browser ‘Comet’

🚀 Perplexity AI is taking its AI-first internet vision further with the development of Comet, an intelligent web browser designed to reinvent the browsing experience.

While full details remain under wraps, Perplexity claims Comet will:

✅ Enhance search with real-time AI assistance
✅ Deliver an ‘agentic’ browsing experience powered by AI
✅ Expand Perplexity’s ecosystem, which already includes an AI research tool, an API, and an Android assistant

Challenges Ahead:

Despite its rapid expansion, Perplexity faces increasing legal scrutiny as it ramps up its monetization efforts through advertising and enterprise offerings. Whether Comet can innovate without facing regulatory pushback remains to be seen.

Luma AI Adds Free AI Audio Generation to DreamMachine Video Tool

Luma AI has upgraded its DreamMachine video platform by adding an AI-powered audio generation feature.

🎵 What’s New?

• Users can generate AI audio with one click or customize soundtracks using text prompts.

• Currently in beta, the feature is free for all users.

• This update integrates AI-driven visuals and sound into a single workflow.

OpenAI Expands Deep Research Access to Plus Users

OpenAI has extended Deep Research access — its AI-driven analysis tool — to ChatGPT Plus, Team, Education, and Enterprise subscribers.

⚡ Deep Research Features:

• Autonomous information synthesis from text, images, and PDFs.

• Powered by a specialized o3 model variant for deeper insights.

• Now accessible to Plus users (10 queries/month) and Pro users (120 queries/month).

OpenAI’s move comes as competitors like DeepSeek and Perplexity push for cheaper, more open AI models, challenging OpenAI’s subscription-based approach.

DeepSeek Fast-Tracks Launch of Next-Gen AI Model, R2

Following the success of DeepSeek R1, the company is now accelerating the release of its R2 model, which was originally planned for May.

💡 What’s Special About R2?

• Improved code generation and multilingual reasoning.

• Gaining traction in China, with government-linked enterprises adopting DeepSeek models at scale.

• Nvidia collaboration optimizing R1 for Blackwell, leading to 25x revenue growth and 20x lower inference costs.

With API access now restored after capacity constraints, DeepSeek is racing to maintain its edge in the rapidly evolving AI landscape.