Bluesky Thread

Claude 4: Sonnet & Opus

May 22, 2025 View original thread

Claude 4: Sonnet & Opus

"GitHub says Claude Sonnet 4 soars in agentic scenarios and will introduce it as the base model for the new coding agent in GitHub Copilot."

www.anthropic.com/news/claude-4

A bar chart titled **"Software engineering"** with the subtitle **"SWE-bench verified"** compares the **accuracy** of different AI models on a software engineering benchmark. The y-axis represents **accuracy percentage**, ranging from 0 to 80.

Each bar represents a different model, and for some models, two values are shown:

* The **darker shaded portion** of the bar shows **verified accuracy**
* The **lighter extension** (where present) shows **accuracy with parallel test-time compute**

**Models and Results:**

1. **Opus 4**

* 72.5% verified accuracy
* 79.4% with parallel test-time compute

2. **Sonnet 4**

* 72.7% verified accuracy
* 80.2% with parallel test-time compute

3. **Sonnet 3.7**

* 62.3% verified accuracy
* 70.3% with parallel test-time compute

4. **OpenAI Codex-1** – 72.1%

5. **OpenAI o3** – 69.1%

6. **OpenAI GPT-4.1** – 54.6%

7. **Gemini 2.5 Pro** – 63.2% (with a footnote: “Preview (05–06)”)

**Visual Style:**

* Opus and Sonnet models are in **reddish-orange** bars
* All other models are in **muted gray**
* The highest value (Sonnet 4 at 80.2%) is emphasized as the chart peak

The chart highlights that Sonnet 4 and Opus 4 outperform other models, especially when parallel test-time compute is used.

21 1

You can now use MCP directly through the API for both OpenAI & Anthropic models. Everything is trending toward agentic

this is incredible. it got *every single turn*. even at the beginning and end. not a single missed turn. is this using google maps???

(sonnet-4.0)

A screenshot of a ChatGPT conversation. The user asks:
**"what's the best way to drive from Raleigh, NC to Greene, NY?"**

ChatGPT responds with a detailed driving route and summary:

---

**The best route from Raleigh, NC to Greene, NY** is typically via I-95 North and I-495 (Capital Beltway), then I-270 North to I-81 North. Here's the most efficient path:

### **Recommended Route:**

1. Take I-40 West from Raleigh to I-85 North
2. Follow I-85 North through Virginia
3. Merge onto I-95 North toward Washington DC
4. Take I-495 West (Capital Beltway) around DC
5. Take I-270 North toward Frederick, MD
6. Merge onto I-81 North through Maryland and Pennsylvania
7. Take I-380 North in Pennsylvania
8. Follow I-81 North to I-88 East in New York
9. Take I-88 East to NY-12 North toward Greene

---

### **Distance & Time:**

Approximately **500–520 miles**, taking about **8–9 hours** of driving time under normal conditions.

pricing

$This is a Claude API pricing chart (as of the latest update) showing input and output token costs (in USD per million tokens) across various generations and models. Here's the breakdown: --- ### **Legacy Models** | Model | Input | Output | | ------------------ | ------ | ------ | | Claude Instant 1.2 | \$0.80 | \$2.40 | | Claude 2 / 2.1 | \$8 | \$24 | --- ### **Claude 3 Family** | Model | Input | Output | | --------------- | ------ | ------ | | Claude 3 Haiku | \$0.25 | \$1.25 | | Claude 3 Sonnet | \$3 | \$15 | | Claude 3 Opus | \$15 | \$75 | --- ### **Claude 3.5 Family** | Model | Input | Output | | ----------------- | ------ | ------ | | Claude 3.5 Haiku | \$0.80 | \$4 | | Claude 3.5 Sonnet | \$3 | \$15 | --- ### **Claude 3.7** | Model | Input | Output | | ----------------- | ----- | ------ | | Claude 3.7 Sonnet | \$3 | \$15 | --- ### **Claude 4 (New)** | Model | Input | Output | | --------------- | ----- | ------ | | Claude 4 Sonnet | \$3 | \$15 | | Claude 4 Opus | \$15 | \$75 | --- The lowest cost model overall is **Claude 3 Haiku**, and the most expensive is **Claude 3/4 Opus**. Claude 3.7 and 4 Sonnet pricing remain unchanged from the 3.5 generation.$

this is not simply good, it's stinkin fast!!

More like this