Claude 4 Sonnet & Opus + Free APIs : RIP Gemini 2.5 Pro! This NEW AI Model is the best AI Coder?

Key Concepts

• Claude 4 Opus and Sonnet: New generation of AI models from Anthropic.
• Extended Thinking: Models can use tools like web search to improve responses.
• Tool Use: Models can use tools in parallel and follow instructions more precisely.
• Memory Capabilities: Improved ability to extract and save key facts over time.
• Code Execution Tool: Allows Claude to run code and use the response in its reasoning.
• MCP Connector: Enables Claude to connect to any remote model context protocol server.
• Files API: Allows users to store and access documents when building with Claude.
• Prompt Catching: Ability to catch prompts for up to 1 hour.
• OnDemand: An all-in-one AI operating system for creating AI agents and AI automation workflows.
• Kilo Code: A merge of Klein and Rue code, offering coding capabilities with free credits.

1. Introduction of Claude 4 Opus and Sonnet

• Anthropic has launched Claude 4 Opus and Claude 4 Sonnet, the next generation of Claude models.
• Claude Opus 4 is positioned as the world's best coding model, excelling in complex, long-running tasks and agent workflows.
• Claude Sonnet 4 is a significant upgrade from Claude Sonnet 3.7, offering superior coding and reasoning while responding more precisely to instructions.

2. New Paradigms: Extended Thinking and Tool Use

• Both models support "extended thinking" with tool use, allowing them to use tools like web search during reasoning.
• This enables Claude to alternate between reasoning and tool use to improve responses.
• The models can use tools in parallel, follow instructions more precisely, and demonstrate improved memory capabilities when given access to local files.
• They can extract and save key facts to maintain continuity and build tacit knowledge over time.

3. General Availability and Collaboration Features

• Claude 4 is now generally available after receiving positive feedback during the research preview.
• Claude Code supports background tasks via GitHub actions and native integrations with VS Code and Jet Brains.
• Edits are displayed directly in files for seamless pair programming.

4. New API Capabilities for AI Agents

• Four new capabilities have been released on the Anthropic API to enable developers to build more powerful AI agents:
  • Code Execution Tool
  • MCP Connector
  • Files API
  • Ability to catch prompts for up to 1 hour

5. Comparison to Previous Models and Pricing

• The new models can perform tasks similar to those of 03 within the chat GPT interface but via API.
• They can use tools and call them in the reasoning traces, similar to 03.
• Both models are hybrid, offering near-instant responses and extended thinking for deeper reasoning.
• Pricing remains consistent with previous Opus and Sonnet models:
  • Opus 4: $15 and $75 per million tokens
  • Sonnet 4: $3 and $15 per million tokens
• Opus is considered "extremely pricey."

6. Model Benchmarks and Performance

• On the swench benchmark, Sonnet performs slightly better than Opus.
• Opus 4 shows a 5-10% improvement over previous models, which is considered "underwhelming."
• Claude Code is now better.

7. Code Execution Tool Details

• The code execution tool allows Claude to run code, see the response, and use it in its reasoning.
• Users receive 50 free hours of usage per day, then pay 5 cents per hour per container for additional usage.
• This tool is not turned on by default in the API.

8. MCP Connector Details

• MCP usage is more streamlined, enabling developers to connect Claude to any remote model context protocol server without writing client code.
• The Anthropic API handles connection management, tool discovery, and error handling automatically.
• Users can add a remote MCP server URL to their API request to access third-party tools.

9. Files API Details

• The Files API allows users to store and access documents when building with Claude.
• Documents can be uploaded once and referenced repeatedly across conversations.
• The Files API integrates with the code execution tool, enabling Claude to access and process uploaded files directly during code execution.
• Claude can produce files such as charts and graphs as part of the response.

10. Extended Prompt Catching

• Users can choose between a standard 5-minute time to live (TTL) for prompt catching or opt for an extended 1-hour TTL at an additional cost.
• The 12x improvement can reduce expenses for long-running agent workflows.

11. Availability and Integrations

• The model is available on the cloud platform and API platform.
• It is not yet available on open router but should be soon.
• Sonnet is available on requesty with a $5 free credit.
• It can be used through Klein and Rue code for coding.
• Once available on open router, it can be used via kilo code with a $20 free credit.
• Claude Code is integrated with VS Code, opening a diff view automatically.
• It is also integrated with GitHub, allowing users to tag Claude in GitHub for autonomous tasks.
• There is a Claude Code SDK available.

12. OnDemand AI Operating System

• OnDemand is an all-in-one AI operating system for creating AI agents and AI automation workflows.
• Users can create agents for social media posting, automated trading, and more.
• It has an agent flow builder for assembling agents, model endpoints, fulfillment prompts, and chat prompts into cohesive workflows.
• Workflows can be scheduled to execute at custom intervals.
• It has an advanced voice mode for making voice agents.
• Outputs can be delivered via email, Slack, or actionable agents.
• Free $50 credits are available for trying it out.

13. Conclusion

The launch of Claude 4 Opus and Sonnet brings new capabilities and improvements to Anthropic's AI models. The extended thinking, tool use, and new API features offer developers more power and flexibility in building AI agents and applications. While the pricing of Opus remains high, the availability of Sonnet and the various integrations with tools like VS Code and GitHub make these models accessible and useful for a wide range of users.