Plotly

Can DeepSeek 4.1 Flash do data analysis? We wired up DeepSeek 4.1 Flash into Plotly Studio Embedded and put it through our data analytics exam. Overall, we find that it's a very capable model. It passes all of the straightforward data analysis questions and handles ambiguity well. It falls short interpreting statistics and finding a "needle in a haystack". It used to be the best cheap model. But recent releases in the last month by Anthropic (Haiku 5.5) and OpenAI (Luna) have reset that narrative. But it remains the cheapest open weight model that we've tested. Read the full deep dive here: https://lnkd.in/g3MxGtjp

Introducing the Plotly Cloud Claude Connector 🎉 This connector can deploy, manage, and diagnose Dash apps on your behalf directly from Claude or Claude Code. Under the hood, it uses Plotly Cloud's MCP URL, which means that this capability is available to any AI product that supports MCP, including ChatGPT and Codex. Our favorite feature is letting Claude instrument and diagnose issues by reading the app logs on Plotly Cloud. Dev != Prod, and this feature allows Claude to go so much further on its own in getting your application out into the world. Read the full post here: https://lnkd.in/gXCMTe2H

Here's an early look of Anthropic Haiku 5.5 (released today!) in Plotly Studio Embedded 🙆♀️ Haiku 5.5 is 9x cheaper than Haiku 4.5 and 2 letter grades better in our data analytics exam. It's also now the fastest model, completing each question in ~14s. Overall very good at data analysis. It got a 100% answering all of the straightforward data analytics questions. It fell short answering some of the questions that required some deeper statistical analysis like looking into other variables. In other words, it's not as persistent as other models in its analysis, which is what you'd expect from a model like this. Similar ballpark to OpenAI Luna in price, cost, and accuracy. These are very cheap models: $0.38 to answer 40 in-depth data analytics questions (compared to $15 for Opus 5.5 or $20 for Astra). Read our full report here: https://lnkd.in/gQ5Jw_Jh

Mistral Large 4 is out! This is a European sovereign open weight model. Today's release is a massive improvement over their last model from April of this year; it's 10x cheaper and performs 2 grades higher on our data analytics benchmark. Congrats to the Mistral team 👏 This benchmark runs an exam of 43 data analytics tasks in the Plotly Studio Embedded harness. The exam runs against a data warehouse with 20M rows, 44 tables, and 554 columns and tests a broad range of data analytic capabilities including answering questions, recognizing ambiguity, avoiding hallucinations, and finding a needle in a haystack. It cost $1.94 in inference tokens for Mistral Large 4 to complete the exam. This is very cost efficient and similar to models like OpenAI GPT-6.1 Sol and Qwen 3.8. It's also fast: it answers each question in about 22s. Overall it is a great model for data analytics, especially for those seeing a European sovereign model.

LLMs → Plotly Chart Spec FTW

Dash is having a major moment in the era of AI in the energy industry. When I first released Dash in 2017, my goal was to enable data scientists and field engineers to get their work out of Python notebooks and antiquated BI tools and into modern web apps that they could easily share and actually run operations on. With the model releases over the last 6 months (and even more so in the last 3 weeks), I’m seeing a massive step change in our customer’s ambitions. Rather than building apps to do the 10% of modeling or analytics that they couldn’t do with preexisting point-and-click tools, they’re rebuilding a good number of pre-existing assets in code with Python and Dash with the goal of decommissioning old dashboards to cut license costs and improve the ergonomics of their internal software. There’s a lot of clunky internal software out there, and they’re giving it new life. Everyone is tinkering with vibe coding apps. I’ve probably built close to 30 apps over the last 12 months. It’s just amazing what you can one-shot with Dash and a frontier model these days. The thing that unlocks AI is getting these apps out of the terminal and onto a company’s server. Normally deploying and managing an internal app would require access to a cloud console, SSH access to a server, or submitting a ticket to IT. But with Dash Enterprise, IT manages the _platform_ and the scientists, engineers, and analysts (and their agents!) have free rein to build, publish, and manage apps on their own. The reason this works is that the platform is run entirely on their infrastructure behind their firewall (not even Plotly can access it), which means they can build truly mission critical, IP and data-sensitive internal tools and apps and not just prototypes. The winning analytics and internal app platform allows you to have full control over the things that differentiate your organization (the data, visualization, models, workflows) while relying on the framework and the platform to handle everything else. You should be responsible for maintaining the minimum amount of code necessary: use coding agents to build a bespoke workflow tool that fits your organization’s processes perfectly, but don’t vibe code the auth flow or the UI state management. That’s what’s guiding our development behind the Dash framework and platform: Build data apps with Python and the Dash framework and then land these apps on the Dash Enterprise platform with SSO, group permissions, [prototype → UAT → production], and a CLI that agents can use to deploy, monitor, and diagnose. To get a sense of what energy teams have been building with Dash Enterprise, check out our new page here: https://lnkd.in/gySHt77B

September brought the new Claude connector for Plotly Cloud, a deep dive into plotly.py 7.0 and plotly.js 4.0, and new blogs from Chris Parmer and Matthew Brown 🚀 Catch up on it all in this month's newsletter.

Exciting news! The Plotly Cloud Connector is now officially available in the Claude Connector marketplace. With a one-click install, now you can use Claude.ai to one-shot an app, publish it, and monitor its health all from Claude's web interface. If you use a different AI provider, you're covered too! Just configure Plotly Cloud's MCP endpoint with your tool of choice and you'll get the same benefits. Check out this blog post or our documentation for more information. - Plotly Cloud Connector: https://lnkd.in/gx5ffXfG - Blog: https://lnkd.in/grwHcH_J - Documentation: https://lnkd.in/g2UjHuHU

Maribick Postanes creating some impressive 2D + 3D plots made with Plotly in the energy space.

OpenAI GPT-6.1 Sol is the second model to get a 100% score on our data analytics benchmark. The first model was GPT-6 Astra. GPT-6.1 Sol passes the exam at nearly 10% of the cost ($2.61 vs $19.97) and about 53% faster than GPT-6 Astra as well. This looks like a very strong model for data analytics and visualization.