OpenAI's fight to win back developers

Presented by:

Welcome to Runtime! Today: OpenAI releases a new model that could dent Anthropic's hold over enterprise software teams, Anthropic reloads to fight back, and the latest enterprise moves.

Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime each week, and if you value independent enterprise tech journalism, click the button below and become a Runtime supporter today.

Become a Runtime Supporter

Looking for a Spark

It's no secret that among the frontier model companies, Anthropic has found more success inside enterprise tech organizations than OpenAI, which otherwise dominates the consumer market for AI tools. It's also no secret that OpenAI needs to find a way to close that gap if it is to even come close to hitting its revenue projections, and Thursday it released a new model that could help.

GPT-5.3-Codex-Spark (just rolls right off the tongue) is a new, smaller version of the GPT-5.3-Codex model OpenAI introduced just last week. The "research preview" release is notable for several reasons, including the speed at which the new model can create code and the chip used to serve up its results.

"Codex-Spark is our first model designed specifically for working with Codex in real-time—making targeted edits, reshaping logic, or refining interfaces and seeing results immediately," the company said in a blog post.
The model can deliver more than 1,000 tokens per second, according to OpenAI, and it is almost as accurate as GPT-5.3-Codex but delivers results in a fraction of the time needed by the full-on 5.3-Codex model, which is still the better bet for more difficult tasks.
That could make it ideal for spinning up side projects to test ideas or experiment with different workflows in an application, the kind of work that would be considered a "nice-to-have" just a few years ago but almost certainly wasn't worth assigning to an expensive engineer.

The Codex-Spark model is also the first OpenAI model to run on Cerebras' AI inference chips, following a deal signed between the two companies last month. While Nvidia still controls the AI chip market — especially when it comes to training — alternatives like Cerebras, Google Cloud's TPUs, and AWS's Trainium chips are making inroads.

Cerebras' Wafer Scale Engine 3 is the chip behind Codex-Spark, which the company describes as "the largest AI chip ever built" and capable of delivering 28 times more computing power than Nvidia's B200, based on its Blackwell architecture.
The Cerebras design also has another advantage over Nvidia's designs, as noted by The New Stack: it uses on-chip memory, which reduces the amount of the increasingly difficult-to-find system memory chips that AI providers would need to buy.
However, OpenAI was very careful to acknowledge Thursday that "GPUs remain foundational across our training and inference pipelines and deliver the most cost effective tokens for broad usage," given reports that OpenAI and Nvidia are having some relationship problems.

It's a little hard to tell how developers have responded to OpenAI's coding-agent era, which kicked off in earnest last May with the launch of its Codex tool. According to CB Insights, "OpenAI does not disclose any information regarding its AI coding product called Codex," which is a curious decision given how closely its rivals are associated with AI coding unless those numbers aren't great.

Software developers have proven themselves quite willing to switch AI coding tools when something new and better comes along, as the swift progression from GitHub Copilot to Cursor to Claude Code shows.
But it seems likely, based on years of history, that enterprises will eventually decide to standardize on a certain set of tools and models for internal development at some point down the road, known as the "golden path" or "paved road."
Now OpenAI has a fresh opportunity to get Codex established inside those companies given that 2026 appears to be shaping up as the year that AI coding agents enter mainstream software development teams.
And if it can figure out how to build a model that works at Codex-Spark speed but produces more accurate results, it might start disclosing its progress.

A MESSAGE FROM GITLAB

Speeding up code isn't enough. GitLab delivers AI across the entire software lifecycle. Learn more at GitLab.com.

Series G Force

At some point over the next 18 months we're going to learn if the billions invested in frontier model companies over the last several years will come off looking shrewd or hilarious. Anthropic added another pile of cash to its bank account Thursday, raising $30 billion at a valuation of $380 billion.

The company also said it is now on pace to record $14 billion a year in revenue on an annualized basis "less than three years since Anthropic earned its first dollar in revenue." It also noted that "business subscriptions to Claude Code have quadrupled since the start of 2026, and enterprise use has grown to represent over half of all Claude Code revenue."

Anthropic has now raised $63.7 billion in funding, and said it would use the new funds to build out an even larger infrastructure footprint. Wall Street expects both OpenAI and Anthropic to test the public-market waters later this year, and if it maintained its $380 billion private valuation in an IPO it would be worth more than IBM, SAP, and Salesforce.

Enterprise moves

Aneel Bhusri is once again the CEO of Workday, replacing Carl Eschenbach, who steps down amid incredible stock-market chaos surrounding SaaS companies.

Chris White is the new chief revenue officer at Camunda, joining the business-process management company after sales leadership roles at Zimperium and Druva.

Stephen Patak is the new chief revenue officer at Blaize, joining the edge computing company after sales leadership roles at Nile and Ubicquia.

Ariel Kelman is the new chief marketing officer at AMD, following similar roles at Salesforce and AWS.

Kelly Morgan is the new chief customer officer at KnowBe4, joining the security company after similar roles at Docusign and Sykes.

Salesforce hired or promoted six new executives in the last few weeks, including new chief security officer Iain Mulholland and new president of enterprise and AI technology Joe Inzerillo, following a wave of executive departures in 2026, according to Business Insider.

The Runtime roundup

Cisco suffered its worst day on the stock market in four years even after reporting solid earnings, which Bloomberg attributed to a weaker-than-expected profit forecast thanks to the memory-chip shortage.

Palo Alto Networks watered down an threat intelligence report that blamed a "state-aligned group that operates out of Asia" for a recent hacking spree instead of China over fears of retaliation, according to Reuters.

Google's AI teams have noticed "an increase in model extraction attempts or 'distillation attacks'" in recent months, with the goal of stealing the intellectual property behind leading AI models.