- The Clever Nest
- Posts
- New Model
Hey clever folks,
Anthropic just released Claude Opus 4 and Claude Sonnet 4 & they’re already redefining the AI leaderboard.
Claude Opus 4: The Best Coding Model Yet
Opus 4 now leads on SWE-bench Verified, hitting 72.5% accuracy in realistic software engineering tasks (and up to 79.4% with test-time compute). It also tops Terminal-bench and handles multi-hour agent workflows without breaking a sweat.
(Talk about being smarter than any human)
Claude Sonnet 4
Sonnet 4 is a major upgrade from 3.7, scoring 72.7% on SWE-bench and showing huge gains in:
Coding
Reasoning
Following complex instructions
Maintaining context across long sessions
GitHub will use Sonnet 4 to power their next Copilot coding agent. (This is set to become your coding copilot.)
Benchmarks: Claude 4 vs OpenAI & Gemini
Claude 4 models outperform almost every other LLM across:
Coding (SWE/Terminal-bench)
Graduate-level reasoning
Visual and multilingual tasks
Claude Code Is Now GA
Claude Code (previously in preview) is now generally available, with:
VS Code & JetBrains integrations
GitHub Actions support
Claude Code SDK to build your own agents
New /install-github-app bot for PRs
Build AI dev tools like:
CI/CD bots
Test fixers
Review assistants
New Dev Capabilities
The Claude API now supports:
Tool use (code execution, web search)
Model Context Protocol (MCP)
Files API
Prompt caching (up to 1 hour)
These additions open the door for more powerful, memory-driven agents. (Imagine, not forgetting any code, pure magic)
Real-World Use: Claude As Your Daily Assistant
In this video, Maggie from Anthropic shows how she uses Claude:
Summarizing her inbox, calendar, Asana
Prepping literature reviews from Drive + web
Auto-generating research insights
Delegating prototype builds with Claude Code
Structuring tasks into Asana via Remote MCP
Opus 4 even builds internal memory like a "Navigation Guide" while playing Pokémon.
Get the latest AI trend delivered straight to your inbox.
— Written by Aaron & The Clever Nest Team