The first entry is always the pin's source. Overall confidence is a weighted average of how firmly each reference supports the start and end times used above; a reference counts half as much for every 180 days older than the newest.
The post is dated "Oct 22, 2024" and says "The upgraded Claude 3.5 Sonnet is now available for all users. Starting today, developers can build with the computer use beta"; TechCrunch[4] and CNBC[5] report the release "on Tuesday" the same day.
Anthropic's[1][3] model card addendum: the upgraded 3.5 Sonnet sets new state-of-the-art results in agentic coding (SWE-bench Verified), agentic tasks (TAU-bench) and computer use from screenshots (OSWorld), with a 14.9% average OSWorld success rate from screenshots alone.
Anthropic's[1][2] research post on the skill: Claude looks at screenshots and "counts how many pixels vertically or horizontally it needs to move a cursor in order to click in the correct place", and was trained on simple software such as a calculator and a text editor without internet access.
TechCrunch, 2024-10-22: "Anthropic[1][2][3] on Tuesday released an upgraded version of its Claude 3.5 Sonnet model that can understand and interact with any desktop app" through a Computer Use API in open beta on the API, Bedrock and Vertex AI.
CNBC, 2024-10-22: chief science officer Jared Kaplan says it can do tasks with "tens or even hundreds of steps"; Amazon had early access and beta testers included Asana, Canva and Notion; released Tuesday in public beta for developers.
Something missing or wrong? Say it in your own words: a link that backs this pin up, a different start or end date and why, or a fact it lacks or gets wrong. The AI checks it against this pin's sources, searches for better ones, and adds any page that backs you up. The pin's own sources still count most. A picture that shows something else, or shows it badly, is looked at too, and moved down or replaced.
From Anthropic's launch post, which gives the benchmark figures in text and states no context window, maximum output or price figures. · source
It worked exactly as advertised.
We only have very early feedback so far, so it's hard to tell how much what I will be calling Claude 3.5.1 improves performance in practice over Claude 3.5. It does seem like it is a clear improvement.
But don't fire your secretary just yet. In an evaluation designed to test an AI agent's ability to help with airline booking tasks, like modifying a flight reservation, the new 3.5 Sonnet managed to complete less than half of the tasks successfully.
The new Claude 3.5 Sonnet model shows improved performance, especially in logical reasoning, mathematical problem-solving and programming tasks.
| Initial explorations of Anthropic's new Computer Use capability · Oct 22, 2024 | 0.60 |
|---|---|
| Anthropic's new AI model can control your PC · Oct 22, 2024 | -0.20 |
| Anthropic releases updated, smarter Claude Haiku 3.5 and Sonnet 3.5 model · Oct 22, 2024 | 0.70 |
| Claude Sonnet 3.5.1 and Haiku 3.5 · Oct 24, 2024 | 0.30 |
| Benchmark | Claude 3.5 Sonnet (new) | Claude 3.5 Haiku | Claude 3.5 Sonnet | GPT-4o* | GPT-4o mini* | Gemini 1.5 Pro | Gemini 1.5 Flash |
|---|---|---|---|---|---|---|---|
| Graduate level reasoning (GPQA Diamond) | 65.0% (0-shot CoT) | 41.6% (0-shot CoT) | 59.4% (0-shot CoT) | 53.6% (0-shot CoT) | 40.2% (0-shot CoT) | 59.1% (0-shot CoT) | 51.0% (0-shot CoT) |
| Undergraduate level knowledge (MMLU Pro) | 78.0% (0-shot CoT) | 65.0% (0-shot CoT) | 75.1% (0-shot CoT) | — | — | 75.8% (0-shot CoT) | 67.3% (0-shot CoT) |
| Code (HumanEval) | 93.7% (0-shot) | 88.1% (0-shot) | 92.0% (0-shot) | 90.2% (0-shot) | 87.2% (0-shot) | — | — |
| Math problem-solving (MATH) | 78.3% (0-shot CoT) | 69.2% (0-shot CoT) | 71.1% (0-shot CoT) | 76.6% (0-shot CoT) | 70.2% (0-shot CoT) | 86.5% (4-shot CoT) | 77.9% (4-shot CoT) |
| High school math competition (AIME 2024) | 16.0% (0-shot CoT) | 5.3% (0-shot CoT) | 9.6% (0-shot CoT) | 9.3% (0-shot CoT) | — | — | — |
| Visual Q/A (MMMU) | 70.4% (0-shot CoT) | — | 68.3% (0-shot CoT) | 69.1% (0-shot CoT) | 59.4% (0-shot CoT) | 65.9% (0-shot CoT) | 62.3% (0-shot CoT) |
| Agentic coding (SWE-bench Verified) | 49.0% | 40.6% | 33.4% | — | — | — | — |
| Agentic tool use (TAU-bench), retail | 69.2% | 51.0% | 62.6% | — | — | — | — |
| Agentic tool use (TAU-bench), airline | 46.0% | 22.8% | 36.0% | — | — | — | — |
* Anthropic's evaluation tables exclude OpenAI's o1 model family, as they depend on extensive pre-response computation time, unlike typical models.