Just last night, the news that Anthropic was going to release a new model began to fester widely in the AI community, but not the expected Claude 4.0, but rather the 3.7 Sonnet version.
Early this morning, Anthropic's new flagship model arrived, as promised, with the official release of its smartest model yet and the first hybrid inference model on the market -- Claude 3.7 Sonnet.
Having learned about it, there are probably two things:
① Mixed Reasoning Model 3.7sonnet
②claude code code tool
First, what is mixed reasoning?
Claude 3.7 Sonnet can generate near-instantaneous responses or show users extended, step-by-step thinking.
According to Anthropic, "One model, two ways to think...", i.e., ordinary and extended thinking modes.
It's both a normal LLM and an inference model. You can choose when you want the model to answer normally and when you want it to think longer before answering.
Here Normal is normal, the normal model is the enhanced version of 3.5, and the extended model (that is, the reasoning model with thinking) is used when reasoning is needed.
In layman's terms, it means thinking when you have to think, and directly outputting the answer when you don't have to think, saving more reasoning costs.
And along with the release of Claude 3.7 Sonnet, Anthropic has also introduced Claude Code, a command-line tool for intelligent coding.
What is the Claude Code?
Simply put, it's an integration tool for code that's currently available as a limited research preview release, enabling developers to delegate a large number of engineering tasks to Claude directly from their endpoints.
How awesome is that?
Direct output of programs for people without programming knowledge, like this one:
One round of dialog directly generates a runnable Minecraft simulation ↓
And then there's this one, where a few rounds of dialog can lead directly to an Apple Watch-based trivia game with heartbeat sensing ↓
All of these codes are generated through dialog, and just 5 prompts required to complete them without a hitch, which is cool for thieves.
Write all the code for the Snake game for the Apple Watch, among other things:
- Use your pulse to determine the speed of the snake, we need to use HealthKit to do this (and tell me how to set it up)
- You swipe on the screen to move the snake up, down, left, right
- Walls don't kill you, you just appear from the other side, so the only way to die is to hit your snake, like the Nokia version
- Using graphics similar to the Nokia version, those screens have a camouflage green look to them
- Write all the code and outline each file so I can copy and paste and run it
Amazing, is all I can say.
And, the author shared the link to this conversation, carrying it here ↓
https://claude.ai/share/ac45287c-3d56-4a7d-a438-bbe8c5e3ce06
This is the power of claude code.
Thanks to this, it's not an exaggeration to say that Claude 3.7 Sonnet is the best coding model to date.
In addition, thecursorIn the midst of that, the 3.7 model has also been updated (the action is really fast).
This is great, really everyone can be a programmer.
As one user commented, "Anthropic makes people smile and get excited with every release! (Mostly, it excites Openai a bit.) It puts cutting-edge reasoning at your fingertips."
Since June 2024, Sonnet has been the model of choice for developers worldwide.
Claude Code was born out of Anthropic's deep exploration of developer coding habits and needs.
Well, here's Claude's official website:
🔗: https://claude.ai/
Free users can experience the regular version of 3.7, and Extended Thinking requires a subscription.
Those who need it can experience the experience.
Of course, the most frustrating thing about Claude is the strict ip-blocking, so it's up to you whether you can log in or not haha~~