Anthropic Company Profile: 75+ Statistics and Trends in Q4 2023
Explore Anthropic's impactful journey in the realm of AI. From the emergence of Claude AI to cutting-edge research, delve into the statistics shaping the future.
Safety Above All: Anthropic’s Uniquely Principled Approach to AI Development
Throughout 2023, one professional field that has exploded in relevance and burst onto everybody’s screens is that of Artificial intelligence (AI). Although much of this attention has been centred on OpenAI and ChatGPT, a derivative large language model (LLM) named Claude AI has drawn serious interest from investors and enthusiasts alike.
The pioneering firm behind Claude is a San Francisco-based startup named Anthropic, which opened its doors in early 2021. So far, Anthropic’s call to fame has been its focus and relentless dedication to developing reliable AI systems that lead the industry in responsibility and safety.
Moreover, Anthropic has published insightful research uncovering the potential impacts of AI on the world and shedding light on any underlying risks and opportunities that come with leveraging the technology.
It is important to note that above all else, Anthropic virtuously commits itself further than the technological innovation typically expected of tech companies, instead setting safety as their success metric. They strive to build AI systems that are honest, helpful, and harmless.
This unique approach has allowed Anthropic to build a systematic pipeline wherein the firm carries out experimental research, applies any findings to its AI offerings, and crowdsources valuable insights from the global community. This community response, in turn, further strengthens the applicability of its AI suite and aligns stakeholders (researchers, data scientists, engineers, policy analysts, etc.).
Read on below to learn more about the statistics which provide more insight and context on Anthropic AI.
Company History (founding, releases)
Even though Anthropic is a relatively young company, it has a rich and interesting history. Most notably, the company was created and is currently headed by AI’s most alluring sibling duo of Dario and Daniela Amodei. The pair, along with other OpenAI alumni, founded Anthropic after not being satisfied with OpenAI’s strategic direction back in 2021. The following milestones show the up to date progression of the budding AI outfit:
Who Founded Anthropic?
Anthropic was founded in 2021 by former OpenAI research executives, including Dario Amodei, Ilya Sutskever, and Greg Brockman (Source)
What AI Models Have Anthropic Released?
Anthropic launched Claude AI (ver 1.0) chatbot on March 14, 2023 (Source)
First major upgrade to Claude 1.0 (now ver 1.3) is made public on April 18, 2023 (Source)
Anthropic releases overhaul to Claude AI, introducing version 2.0 on July 11, 2023 (Source)
Latest iteration of Claude AI (Ver 2.1) is made available on November 21, 2023 (Source)
On March 4th, 2024, Anthropic introduced a new AI model — Claude 3, which sets a “new industry benchmark” across a variety of cognitive tasks. (Source)
Following the release of Claude 3, Anthropic launched Claude 3.5 Sonnet in June of 2024. (Source)
Development
Aside from the technical developments made with Claude AI, the following milestones below indicate the progress made by Anthropic as an AI firm:
What is on the Horizon for Anthropic?
After the release of Claude 3, Anthropic deepens its commitment to creating a revolutionary AI model. (Source)
Amazon concludes a hefty $4 billion investment in Anthropic right before the release of Claude AI’s third model in March 2024 (Source)
Customers and companies of all sizes are using Claude through Amazon Bedrock to reinvent their business strategies and improve user experience. (Source)
Has Anthropic Published any Academic Works?
Anthropic has also published over 15 safety research papers on topics such as constitutional AI, societal impacts, interpretability, red teaming, and scaling laws. (Source)
How Does Anthropic Develop Claude AI?
Claude AI was trained on the combination of two techniques – Constitutional AI and RLHF (Reinforcement Learning from Human Feedback) (Source)
On June 21, 2024, Anthropic announces its first AI model release in the new Claude 3.5 family. Praised to “raise the bar” of the AI industry and “elevate” generative AI like never before, the Claude 3.5 Sonnet may outperform many competitors with up to a 200K token context window.
The model is now available for free on “Claude.ai” or through the Claude iOS application. Paid users can also access the model via “Claude Pro” and “Team” plans, which offer higher limits on rates.
The model is also available through Anthropic’s API, Google Cloud’s Vertex AI, or Amazon Bedrock.
Here’s a detailed overview of the AI model’s specifications:
Cutting Edge Intelligence at Double Speed
Anthropic’s newest Claude 3.5 model sets “new industry benchmarks” for reasoning at the graduate level through the GPQA test and undergraduate knowledge via the MMLU test. It has also undergone a coding proficiency test called “HumanEval,” raising the bar higher for other generative AI models. (Source)
Claude 3.5 operates twice as fast as Claude 3 Opus. As a result, Claude 3.5 is a fantastic option for complex tasks, such as organizing workflow instructions or generating ‘context-sensitive customer support.’ (Source)
Anthropic’s new AI model has solved 64% of problems provided during an ‘internal agentic coding evaluation,’ outperforming the company’s previous model, Claude 3 Opus, which could solve just 38% of the problems. (Source)
Anthropic’s evaluation tests the model’s capability and efficiency in solving bugs or adding functionality to an open-source codebase when it is provided with a natural language description of the required improvements. (Source)
When Claude 3.5 is provided with the relevant tools, it has exceptional coding capabilities. It can create, write, edit, and execute code, applying sophisticated levels of reasoning. In particular, one of Claude 3.5 Sonnet’s leading specialties is the ability to analyze and handle complex coding structures. It can update legacy applications and migrate codebases with ease. (Source)
Claude 3.5 Sonnet Tests: Is It Better Than Other Models?
In a Graduate-level reasoning test (GPQA), Claude 3.5 Sonnet scored over 59.4%, beating Claude 3 Opus by exactly 9% and GPT-4o by about 5%. (Source)
In Undergraduate level knowledge (MMLU), Claude 3.5 had a result of 88.7% on 5-shot and 88.3% on o-shot CoT. Claude 3 Opus’s results were similar, finishing with 86.8% on 5-shot and 85.7% on 0-shot. GPT-4o’s results on 5-shot were not included in the data; however, GPT-4o exceeded Claude 3.5 Sonnet by 0.4% on 0-shot CoT. Google Gemini 1.5 Pro scored over 85.9% on the 5-shot, and Llama-400b had a result of 86.1% (in comparison to Claude 3.5 Sonnet’s comparatively higher 88.7% on 5-shot). (Source)
The results of the coding test (HumanEval) were in Claude 3.5 Sonnet’s favor, scoring 92.0% and beating Claude Opus by approximately 7% and GPT-4o by 1.8%. Gemini 1.5 Pro and Llama 3 400b both had a score of 84.1% (8% lower than Claude 3.5 Sonnet). (Source)
Claude 3.5 Sonnet takes the lead in multilingual math (MGSM) by scoring 91.6% on the 0-shot CoT, and surpassing Claude 3 Opus, as well as GPT-4o, Gemini 1.5 Pro and Llama 3. (Source)
The reading comprehension (DROP) score of Claude 3.5 Sonnet is 87.1, beating Claude 3 Opus (83.1) and GPT-4o (83.4). (Source)
Claude 3.5 Sonnet scored 93.1% in the mixed evaluations test (BIG-Bench Hard), whereas Claude 3 Opus tested at 86.8%. Gemini 1.5 Pro scored 89.2% on the 3-shot CoT and Llama 3-400b had a score of 85.3%. (Source)
In Math problem-solving and grade school math (GSM8K), Claude 3.5 had a score of 96.4%. Claude 3 Opus scored 95.0%, close behind Claude 3.5 in results. (Source)
In visual question answering (MMMU), Claude 3.5 scored 68.3%, ahead of Claude Opus at 59.4%. On the same test, GPT-4o held the highest score of 69.1% and Gemini 1.5 Pro had a slightly lower result of 62.6%. (Source)
In the mathematics test (MathVista) Claude 3.5 performed the best of the models tested. Claude 3.5 produced a score of 67.7%, whereas GPT-4o and Claude Opus lagged behind at 63.8% and 50.5% respectively. (Source)
Claude 3.5 Sonnet also excels with science diagrams (AI2D) with the highest testing score of 94.7%. In contrast, Claude 3 Opus and GPT-4o had results of 88.1% and 94.2% respectively. (Source)
In chart understanding (ChartQA), Sonnet scored 90.8%, beating Opus and GPT-4o, which scored 80.8% and 85.7% respectively. (Source)
On document understanding (DocVQA) Claude 3.5 scored 95.2%, beating Claude 3 Opus’s score of 89.3% and GPT-4o’s score of 92.8%. (Source)
What’s New With Claude - Artifacts Overview
Anthropic states that a “new way to use Claude” is now possible through their newest Artifacts feature, expanding application possibilities through the generative AI tool. (Source)
Whenever the AI model is prompted to generate any type of code, artifacts appear in a ‘dedicated window’ beside the conversation prompt. This encourages a “dynamic workspace” to alter, edit, and view the code’s results in real time. (Source)
The feature allows users to align the code’s output to their preference without having to re-enter the generative prompt — a game-changing and revolutionary new feature in the development of AI tools. (Source)
Anthropic also states that the artifacts feature is “just the beginning” of a broader vision for the Claude model, which will soon become available to support collaboration for teams. (Source)
Is Claude 3.5 Sonnet Secure?
Through “rigorous testing,” Anthropic states that their models are trained to “reduce misuse.” (Source)
On Anthropic’s “Responsible Capability Scaling.” Claude 3.5 Sonnet falls under the ASL-2 category. The scale measures the relative safety level of an AI model. The ASL-2 category is described by Anthropic as showing “early signs of dangerous capabilities… but where the information is not yet useful due to insufficient reliability or not providing information that e.g. a search engine couldn’t.” (Source)
Anthropic has prioritized enhancing safety mechanisms by engaging with external experts for optimization. (Source)
As part of their dedication to safety, Anthropic voluntarily provided the UK Artificial Intelligence Safety Institute (UK AISI) with the Claude 3.5 Sonnet model before they launched it to evaluate its safety. (Source)
The company has also integrated policy feedback from “subject matter experts” to align safety measures and ensure they take into account new or emerging trends for misuse. (Source)
What’s Next for Claude 3.5?
Anthropic’s aim is to improve the correlation between the intelligence, pricing, and speed of their AI models every several months. To have a complete Claude 3.5 family, Anthropic is set to release a 3.5 version of Claude Haiku and a 3.5 version of Claude Opus, later in 2024. (Source)
Anthropic has also been focusing on the development of new modalities and a handful of features to support “more use cases” for organizations and businesses. (Source)
You can now submit feedback on Claude 3.5’s performance directly to Anthropic in-product to shape the “development roadmap” of the model and help to improve your user experience. (Source)
Funding
As an AI darling with a team full of silicon valley pedigree, Anthropic has been largely successful in converting investor interest into large seed investments. The following of which are detailed below:
How Much did Anthropic Raise after Series A, B, C, D and E Funding Rounds?
The company has raised $1.5 billion (July 11, 2023 6:30 AM) in funding to date from investors including Google, Salesforce Ventures, Spark Capital, Sound Ventures and Zoom Ventures (Source)
How Much Did Anthropic Raise in Series A Funding?
Anthropic raises $124M in Series A funding (May 28, 2021) (Source)
Series A led by Jaan Tallinn funding from James McClave, Dustin Moskovitz, the Center for Emerging Risk Research, Eric Schmidt, and others. (May 28, 2021) (Source)
How Much Did Anthropic Raise in Series B Funding?
Anthropic raises $580M in Series B funding (4/29/2022) (Source)
Google invests $300M in generative AI startup Anthropic, corporate funding round (2/5/2023) (Source)
How Much Did Anthropic Raise in Series C Funding?
Anthropic raises $450M in Series C funding (5/25/2023) (Source)
Series C funding led by Spark Capital with participation from Google, Salesforce Ventures, Sound Ventures, Zoom Ventures, and others. (5/23/2023)
How Much Did Anthropic Raise in Series D Funding?
Anthropic raises more than $5M in Series D funding from MIS, Qualcomm, and Intuit in late January 2024 for general technological improvement. (Source)
How Much Did Anthropic Raise in Series E Funding?
One of Anthropic’s largest company investments so far took place on March 27 this year (2024) by Amazon valued at more than $2.75 Billion. The investment has a purpose to integrate Anthropic’s generative AI technology into the selling lines of Amazon. (Source)
Has Anthropic Raised Funds from Corporate Rounds?
SAP invests directly in three AI startups: Cohere, Anthropic and Aleph Alpha (undisclosed amount) -7/18/2023 (Source)
South Korea's SK Telecom has agreed to invest $100 million in generative AI company Anthropic and partner with the startup to develop a multilingual large language model (8/15/2023) (Source)
In a joint interview, the CEOs of Amazon's cloud division and Anthropic said the immediate investment will be $1.25 billion (9/29/2023) (Source)
Google has agreed to invest up to $2 billion in the artificial intelligence company Anthropic, a spokesperson for the startup said on Friday. The company has invested $500 million upfront (10/27/2023) (Source)
What is Anthropic’s Market Value?
Anthropic’s latest valuation as of May 2024, places the market value at approximately $18.4 billion. (Source)
Anthropic listed as 11th most valuable venture-backed pre-IPO (initial public offering) company globally (updated Dec 2023) (Source).
Is Anthropic Looking for New Investors?
Anthropic commencing a new funding round (led by Menlo Ventures) for $750M, nearing a valuation of $18.4B (Source).
Financials
The following data points paint a general picture of the current financial outlook of Anthropic provide a look into the company’s future prospects:
Anthropic predicts to be making around $17 million in monthly revenue by the end of 2023 (Source)
hopes to generate around $500 million annually by the end of 2024. (Source)
Anthropic AI valuation is $18.4 billion (2024). This makes Anthropic one of the most valuable AI safety companies in the world. (Source)
With more than $850 million in annualized revenue forecast for 2024, some sources say that Anthropic could earn as much as $83 million per month. (Source)
Anthropic has informed some investors that it has a $100 million annualized revenue rate—suggesting it has generated more than $8 million in revenue per month. (Source)
Estimate Revenue of $850M by the end of 2024 (Source)
Management and Employees
How Many Employees does Anthropic Employ?
To date, Crunchbase notes that Anthropic has 101-250 employees. (Source)
What is the Average Salary at Anthropic?
Anthropic offers about 296K annually to be a software engineer and more than 628K to be a leading business analyst. (Source)
What Benefits are offered to Anthropic Employees?
Optional equity donation matching at a 3:1 ratio, up to 50% of the equity grant. (Source)
A 401(k) plan with a 4% matching contribution (Source)
Relocation support for those moving to the Bay Area (Source)
Corporate Partnerships and Location
Amazon Bedrock
Quora Poe
Google
SAP
SK Telecom in South Korea
Where is Anthropic HQ Located?
Anthropic headquarters is currently located in San Francisco, California, United States (Source)
Anthropic’s main office was previously Slack’s former HQ location (Source)
Does Anthropic have a Presence Outside the U.S?
In late May(2023) Rishi Sunak personally met with the CEOs of OpenAI, Google DeepMind, and Anthropic (Source)
In May 2023, Anthropic opened an office in London, UK (Source)
As of May 14, Anthropic’s trusted AI assistant Claude 3, was made available to many people and businesses within Europe. (Source)
Anthropic’s Claude AI vs Originality.ai
Owned by Anthropic, Claude is a next-gen generative AI tool, with the capability to understand and answer in a human-like language, making it an ideal chatbot for most users. While most compare it to ChatGPT, Claude excels in refined human responses.
To accurately determine Claude’s effectiveness, we’ll be testing it against Originality.ai’s AI detection tool. We’ll test Claude with a variety of prompts, each aiming to extract the most human output possible.
We will be using Originality.ai to test Claude and the most recent version of the AI chatbot, Claude 3.5 Sonnet.
So, let’s begin with the first tests and uncover how effective Claude is at concealing AI-generated content!
[Test #1] General AI-Generated Text On Any Topic
Let’s start with a simple long-form article to test the Originality.ai AI Detector. During the first tests, we’ll analyze how good Claude is at generating proper human-sounding content, with no extra instructions towards structure and refinement included.
[Prompt #1] - Write a short article (500-1000 words) on the expansion of 5G networks in 2024.
From the 558 words we’ve scanned and compared with Originality.ai, the results are solid:
Claude was not able to conceal the content. The Originality.ai AI detector identifies the content as likely AI with 100% confidence.
For the second prompt, we’ll include extra instructions to make the text sound as unique as possible.
[Prompt #2] - Write a short article (500-1000 words) on the expansion of 5G networks in 2024. Stick to the conventional tone of a human conversation and make the text sound as unique as possible. Refrain from using “robotic” phrases and sentences.
We’ve received a 646-word output from Claude 3.5 Sonnet and run it through Originality.ai’s AI detector. Let’s see the results from the second prompt:
Once again, Originality.ai shows a strong performance with 100% confidence the text is likely AI.
Now, let’s proceed with the second stage of text prompts and provide Claude with a human-written article example to extract ideas and humanize the content as much as possible.
[Test #2] AI-Generated Text Via Human-Written Example
We’ll use identical prompts as with the first two tests but provide Claude with a unique human-written article example that hasn’t been published on the web. This will determine whether the bot can adapt to new writing ideas. Then, we’ll run it through Originality.ai’s AI Detector to see if it can bypass AI detection systems.
[Prompt #1] - Write a short article (500-1000 words) on the expansion of 5G networks in 2024. Use this *article* as an example to style and structure the new piece.
Let’s compare the new output from Claude in Originality.ai:
This time, we received a 619-word article that followed the style of the example we’ve provided. However, despite additional instructions, Originality.ai continues to show excellent capabilities, detecting the content as likely AI with 100% confidence.
Now, let’s provide Claude with an article example and the additional instructions from the second prompt during the previous tests to determine if there’s an improvement.
[Prompt #2] - Write a short article (500-1000 words) on the expansion of 5G networks in 2024. Stick to the conventional tone of a human conversation and make the text sound as unique as possible. Refrain from using “robotic” phrases and sentences. Use this *article* as an example to style and structure the new piece.
Here are the results from the 719 words we’ve scanned:
Once again, Originality.ai performed exceptionally and recognized the final piece as likely AI-generated with 100% confidence. Even with the latest improvements, Originality.ai continued to detect text generated by Claude 3.5 Sonnet!
Anthropic.com Traffic and Reach
Alongside it AI model Claude AI [www.claude.ai], anthropic.com has drawn much interest from the global tech community, currently the site has achieved (Source):
Anthropic.com averages more than 7.8M visits as of data gathered up to September 2024, making it the 8,732th most popular website in the U.S and the 9,491th most visited worldwide.’
The bounce rate of Anthropic.com as of September 2024 is 53.91%
2.72 pages per visit as of September 2024
00:02:25 average visit duration as of September 2024
Most of Anthropic’s traffic is generated from the United States (about 20.7%.) (Source)
Here are which countries visit Anthropic.com the most according to a recent report in September 2024:
United States - 20.7%
China - 15.47%
Korea - 4.54%
India - 4.02%
United Kingdom - 2.88%
Rest of the world - 52.38%
Conclusion paragraph
Considering their strong 2023 where all traffic metrics and consumer interest are trajecting upwards, Anthropic is poised to have a seminal 2024. By retaining its focus on producing the most helpful, honest, and harmless AI solution on the market, Anthropic will surely differentiate itself from competitors like ChatGPT or Google Bard. Along with its strong backing through equity-backed partnerships with technological giants Amazon and SAP, Anthropic has impressed with stark improvements in the technical capabilities of its AI model, Claude. This is all to say that what lies ahead for Anthropic in the coming year will certainly continue to transform the state of AI and keep the rest of the industry on its toes.
Founder / CEO of Originality.ai I have been involved in the SEO and Content Marketing world for over a decade. My career started with a portfolio of content sites, recently I sold 2 content marketing agencies and I am the Co-Founder of MotionInvest.com, the leading place to buy and sell content websites. Through these experiences I understand what web publishers need when it comes to verifying content is original. I am not For or Against AI content, I think it has a place in everyones content strategy. However, I believe you as the publisher should be the one making the decision on when to use AI content. Our Originality checking tool has been built with serious web publishers in mind!
