[#10] Agents Everywhere: June's AI Revolution in Tools, Models, and Billion-Dollar Bets

In partnership with

Hey Geeks,

The AI landscape has shifted dramatically in just the last three weeks. We're seeing a clear pattern emerge: the age of AI agents has officially arrived, with every major player launching tools that can actually do things for you, not just chat.

Let's dive into what's happening and why it matters.

Meta's Llama 4 Maverick: The 400B-Parameter Beast Breaking Speed Records

Meta has officially released Llama 4 Maverick, their massive 400B-parameter model, and it's redefining what's possible in AI inference speed. Unlike previous large models that sacrifice speed for capability, Maverick delivers both – with benchmark tests showing it can generate over 2,500 tokens per second on Cerebras hardware, more than doubling NVIDIA Blackwell's 1,038 tokens per second.

The key difference? Maverick is part of Meta's new multimodal AI lineup that includes text, image, and audio capabilities. It's the largest in the Llama 4 family, which also includes the smaller 17B-parameter Scout model. A third, even more powerful model called "Behemoth" has been delayed due to concerns about its capabilities.

What makes this release particularly interesting is how it's changing the AI hardware landscape. Cerebras has used Maverick to demonstrate that their specialized AI chips can significantly outperform NVIDIA's flagship Blackwell GPUs – with other providers like SambaNova (794 t/s), Groq (549 t/s), Amazon (290 t/s), Google (125 t/s), and Microsoft Azure (54 t/s) trailing behind.

Pricing: Available through various cloud providers with different pricing models. Amazon Bedrock offers it as a fully managed, serverless option, while Groq has exclusive access in Saudi Arabia. Meta plans to offer the models through their own API service soon.

Google's I/O Bombshell: 100 AI Announcements in One Day

Google I/O 2025 was essentially "AI everything" with over 100 AI-related announcements. The most significant:

1. Gemini 2.5 Pro & Flash: Google's latest models now lead the WebDev Arena and LMArena leaderboards. The Flash variant offers stronger performance on coding and complex reasoning while optimized for speed.

2. Agent Mode in Gemini: Similar to OpenAI's approach, this experimental feature lets you describe your end goal and Gemini handles the rest. Coming soon to Google AI Ultra subscribers.

3. Project Mariner's Computer Use: Google is bringing agentic capabilities to the Gemini API and Vertex AI, allowing AI to use computers on your behalf. Companies like Automation Anywhere and UiPath are already exploring its potential.

4. Gemini Diffusion: A new research model that generates text or code by converting random noise into coherent output, similar to how image diffusion models work.

5. MCP Native Support: Google added native SDK support for Model Context Protocol definitions in the Gemini API, making it easier to integrate with open-source tools.

Pricing: Gemini 2.5 Flash is available to everyone in the Gemini app. API pricing remains at 3.50 per million input tokens and 3.50 per million input tokens and 3.50 per million input tokens and 10.50 per million output tokens for Pro.

Find out why 1M+ professionals read Superhuman AI daily.

In 2 years you will be working for AI

Or an AI will be working for you

Here's how you can future-proof yourself:

1. Join the Superhuman AI newsletter – read by 1M+ people at top companies

2. Master AI tools, tutorials, and news in just 3 minutes a day

3. Become 10X more productive using AI

Join 1,000,000+ pros at companies like Google, Meta, and Amazon that are using AI to get ahead.

Sign up and start learning AI

Mistral's Agents API: The New Standard for Building AI Assistants

Mistral just dropped their Agents API, and it's a game-changer for anyone building AI applications. Unlike previous offerings that required complex orchestration, Mistral's approach includes built-in connectors for code execution, web search, image generation, and MCP tools right out of the box.

What makes this special is the persistent memory for conversations (your agent actually remembers context over time) and support for conversation branching. But the killer feature? Dynamic orchestration of multiple agents working together to solve complex problems.

This means you can have specialized agents that handle different parts of a workflow, adding or removing them as needed. For developers, this dramatically simplifies building useful AI applications that can actually accomplish real tasks.

Pricing: Mistral offers a free tier for experimentation, with production pricing at 0.002/1K input tokens and 0.002/1K input tokens and 0.002/1K input tokens and 0.006/1K output tokens – significantly undercutting OpenAI's rates while offering comparable capabilities.

Claude 4 Models: Anthropic's Answer to the Agent Race

Anthropic wasn't about to be left behind. They've launched Claude Opus 4 and Claude Sonnet 4, both capable of undertaking long-running tasks and working continuously for several hours.

Claude Opus 4 excels at coding and complex problem-solving, while Sonnet 4 improves on Sonnet 3.7 with better balance between performance and efficiency. The company also revealed a beta for extended thinking with tool use, parallel tool usage, and general availability of Claude Code.

The most interesting addition to the Anthropic API is the four new capabilities: code execution tool, MCP connector, Files API, and the ability to cache prompts for up to one hour.

The Best AI Models Right Now (June 2025)

The AI model landscape has shifted dramatically in the past month. Here's my current ranking based on extensive testing:

1. Claude 4 Opus - Best for: Enterprise-grade coding & research

   1. Pricing: $20-200/mo

   2. Key strengths: Hours-long sessions, top-tier coding, advanced reasoning

2. Gemini 2.5 Pro - Best for: Huge-context multimodal analysis

   1. Pricing: $19.99-249.99/mo

   2. Key strengths: 1M-token context, multimodal capabilities, Google Cloud integration

3. GPT-4o - Best for: Real-time multimodal chat

   1. Pricing: $0-200/mo

   2. Key strengths: Text-image-audio I/O, 128K context, low latency

4. OpenAI o3 - Best for: Deep chain-of-thought reasoning

   1. Pricing: $0-200/mo

   2. Key strengths: Autonomous tools, coding/math strength, free tier access

5. Claude 4 Sonnet - Best for: Budget-friendly coding assistant

   1. Pricing: $0-200/mo

   2. Key strengths: Fast replies, strong code generation, low API cost

Major Funding & Industry Moves: Billion-Dollar AI Bets Continue

The funding landscape remains red-hot for AI, with several massive rounds closing in the past few weeks:

1. Grammarly - $1 Billion

   1. From General Catalyst's Customer Value Fund

   2. Plans to scale sales/marketing and make strategic acquisitions

2. Neuralink - $650 Million

   1. Investors include ARK Invest, DFJ Growth, Founders Fund, G42, Human Capital, Lightspeed, QIA, Sequoia

   2. Coincides with expansion of clinical trials for brain implant device

3. ClickHouse - $350 Million

   1. Series C led by Khosla Ventures

   2. Also secured $100M credit facility from Stifel Bank and Goldman Sachs

4. Snorkel AI - $100 Million

   1. Series D at $1.3B valuation

   2. Makes tools for evaluation and tuning of specialized AI systems

Other interesting developments:

• Bezos Earth Fund launched its AI for Climate and Nature initiative, with 24 grantees each receiving $50,000 to build climate-focused AI applications.

• Israeli tech startups raised over $950 million across 18 deals in May 2025, with notable mega rounds by AI21 Labs and Classiq.

• IndiaAI provided 200 crore worth of GPUs to Sarvam AI, continuing the government's push to establish India as an AI powerhouse.

Quick Hits:

•API.market continues its impressive growth, now reaching over 5,000 total users. Our in-house Faceswap Image & Video model remains state-of-the-art, with two models now licensed to customers at $600+ USD per month.

•Snyk launched an AI agent security platform called "Snyk AI Trust Platform" designed to help software development teams mitigate business risk when working with AI.

•New Relic created an integration with GitHub Copilot's coding agent that monitors code deployments and automatically detects issues, creating GitHub issues with relevant context.

•DataRobot launched "syftr," an open-source framework for agentic AI that helps developers discover and implement the best combination of components for agentic systems.

•Cast AI introduced Database Optimizer (DBO), which uses intelligent caching to improve cloud database performance through an AI agent that runs a fully autonomous caching layer.

What's Next?

The shift toward agentic AI is happening faster than I expected. Every major player is now focused on building systems that can take action in the real world, not just generate content. The Model Context Protocol (MCP) is emerging as a critical standard for interoperability between these systems.

For developers, this means it's time to start thinking about how to incorporate these capabilities into your products. For businesses, the ROI on AI investments is becoming much clearer as these tools move from experimental to practical.

I'll be diving deeper into specific use cases for these new agent capabilities in next week's newsletter. If there's a particular aspect you'd like me to explore, just hit reply and let me know.

Until next time,

Shashank Agarwal
Founder/CEO
API.market/Noveum.ai