Is OpenAI winning coders back from Anthropic?

Last updated: 21 October 2026
Cover of NewMarketPitch's Anthropic Today: Reality Check report Claude Code or Codex: who wins the coders? Claude Code, on what can be measured: more than three times Codex's traffic among coding apps on OpenRouter. The two take about 70% of company spending on AI coding tools. Chart from NewMarketPitch's Frontier AI Labs report.Who is really ahead in frontier AI, OpenAI or Anthropic? Anthropic, on 3 lines of 5: $17B of third-quarter sales against about $10B, and the top model on the experts' index. OpenAI keeps the crowd and has raised more money. Chart from NewMarketPitch's Frontier AI Labs report.OpenAI or Anthropic: who is closer to making a profit? Anthropic, by its own figures: an $81B sales pace and a $2.5B adjusted profit last quarter. OpenAI sells about $50B a year and forecasts about $280B of cash burn to 2030. Chart from NewMarketPitch's Frontier AI Labs report.Which AI lab's sales grow the fastest? Anthropic, among labs that report their sales: its third-quarter sales were 12 times a year earlier. Mistral's pace grew faster, but from about $20M. Chart from NewMarketPitch's Frontier AI Labs report.Who is fighting whom in frontier AI? OpenAI leads Google on users, Anthropic leads OpenAI on sales, and Alibaba leads DeepSeek on downloads. Alibaba and Kling split video, SSI and Thinking Machines split money and product. Chart from NewMarketPitch's Frontier AI Labs report.Who has held the top AI model, month by month? Anthropic and Google, taking turns: over the last 12 months Anthropic topped the LMArena users' vote for 7 months and Google for 5. OpenAI never did. Chart from NewMarketPitch's Frontier AI Labs report.Is Anthropic catching up with OpenAI in fundraising? Fast. Anthropic raised about $95B in 2026 alone and has now raised about two thirds of what OpenAI has, up from about half at the end of 2025. Chart from NewMarketPitch's Frontier AI Labs report.ChatGPT, Gemini, Meta AI, Claude: who has the most users? ChatGPT, with 1.2 billion people a week. Gemini and Meta AI each claim about a billion a month, and Anthropic gives no count for Claude. Chart from NewMarketPitch's Frontier AI Labs report.

Every funding round, every investor, and what the labs promised in return, in one document. Have a look ↓

Summary

OpenAI is winning coders back from Anthropic in attention and willingness to experiment, but not yet in their primary coding workflows. Codex is growing rapidly, while Claude Code remains the more widely adopted dedicated coding agent among professional developers.

The comeback is substantial. JetBrains found that Codex workplace adoption climbed from 3% to 16% during the first part of 2026, while Claude Code rose from 18% to 39%. Codex grew more than fivefold, yet Anthropic's absolute adoption lead actually widened from 15 to 23 percentage points. Codex growing fivefold while Anthropic's lead widens is why we cover who is ahead and who is falling behind in our report on who's ahead in frontier AI labs.

That apparent contradiction captures the competition. OpenAI is catching up relative to its starting position, but Anthropic is still adding more developer adoption points. The smaller competitor can grow faster proportionally without becoming the market leader.

The strongest evidence against a mass migration is that developers increasingly use several agents. Stack Overflow found 65.5% past-year usage for Claude Code and 29.5% for Codex, with substantial overlap between tools. Codex adoption therefore cannot be treated as proof that developers are abandoning Claude.

OpenAI has nevertheless built a serious user base. Codex surpassed five million weekly active users by June, although approximately 20% were knowledge workers rather than traditional programmers. Its reach is expanding beyond the developers who originally made Claude Code successful.

The technical gap has narrowed considerably. GPT-6 Astra outperformed leading Claude models on Terminal-Bench 4.0 and narrowly led on DeepSWE, while Claude retained advantages on other coding evaluations. There is no longer a straightforward winner across every demanding programming task.

Economics could become OpenAI's biggest weapon. Under OpenAI's published Terminal-Bench conditions, Astra delivered a slightly higher score than Claude Fable 5.1 at approximately 63% lower cost per task. For businesses running large volumes of automated coding work, that difference could outweigh small changes in benchmark rankings.

Codex is also attracting developers who delegate unusually large amounts of programming to agents. JetBrains found that 42% of Codex users generated more than 80% of their code with agents, suggesting OpenAI is gaining traction among people pushing agent-first development particularly hard.

Anthropic's commercial position is harder to challenge than a product ranking suggests. Claude Code exceeded $2.5 billion in annualized revenue run rate, with enterprise customers contributing more than half. OpenAI has disclosed impressive Codex usage figures but no equivalent standalone revenue number.

OpenAI's distribution advantage is enormous: more than one billion weekly users across its products creates a ready audience for Codex. But attracting trials from existing ChatGPT customers is easier than persuading committed Claude Code developers to change their daily workflow. The real test will be sustained primary-tool preference, retention, and enterprise renewals, where Anthropic still has the stronger publicly documented position.

Written by the New Market Pitch teamWe track new markets so founders and investors can move faster. About us

Claude Code or Codex: who wins the coders? Claude Code, on what can be measured: more than three times Codex's traffic among coding apps on OpenRouter. The two take about 70% of company spending on AI coding tools. Chart from NewMarketPitch's Frontier AI Labs report.
Anthropic Today reportThis page and 70+ morePDF Get the report

This is just a preview page. Don’t worry if a number looks old: the report gets refreshed, and you’ll always get the latest version.

Why are developers reconsidering OpenAI after switching to Claude?

OpenAI is making a serious comeback among software developers, but Claude Code is still ahead, and Anthropic is gaining users too.

The shift started when Claude Code became one of the first widely adopted tools that could handle substantial programming tasks directly inside a developer's workflow. Instead of asking a chatbot for snippets, developers could let an agent inspect repositories, edit files, run commands, and fix problems across an entire project.

OpenAI responded by improving Codex, introducing a desktop application, expanding integrations, and releasing much stronger coding models. The results are already visible. In the 2026 JetBrains Developer Ecosystem Survey, Codex adoption among professional developers rose from 3% in January to 16% during May–July. Claude Code climbed from 18% to 39% over the same period. Claude Code going from 18% to 39% in months shows how much has moved in a year, and we follow that in our frontier AI labs report.

OpenAI has made up considerable ground relative to its starting position, even though Anthropic gained more users in absolute terms. The interesting question now is whether Codex's rapid adoption represents developers returning from Claude Code or simply more people adding another AI coding tool.

Did developers really abandon OpenAI for Anthropic?

Anthropic successfully pulled professional developers toward Claude Code, but the available evidence doesn't show that most of them abandoned OpenAI altogether.

Claude Code launched publicly in May 2025 and quickly gained traction among developers working with large repositories, command-line tools, and complex programming tasks. Its appeal was straightforward: developers could delegate actual work instead of repeatedly copying code between an editor and a chatbot.

JetBrains measured Claude Code usage at approximately 3% of professional developers in April–June 2025. By January 2026, adoption had reached 18%, before climbing even further during the following months.

That was an extraordinary rise for a relatively new developer product. Meanwhile, GitHub Copilot, one of the established leaders in AI-assisted coding, saw workplace adoption decline from 29% in 2025 to 21% in the latest JetBrains survey.

There's a distinction worth making here. ChatGPT is OpenAI's general-purpose assistant, while Codex is its dedicated coding agent. A developer could keep using ChatGPT for everyday questions while switching to Claude Code for serious programming.

Anthropic clearly took over much of the coding-agent conversation. What we cannot establish is how many Claude Code users previously relied on an OpenAI coding tool.

Is OpenAI Codex actually catching up with Claude Code?

Codex is closing the relative adoption gap with Claude Code much faster than many developers expected, although Anthropic still has a comfortable lead.

JetBrains' 2026 research provides a useful comparison. Codex increased its workplace adoption by 13 percentage points between January and May–July, representing approximately 5.3 times its initial adoption level. Claude Code added 21 percentage points, or roughly 2.2 times its January level.

At the beginning of the period, Claude Code was used by six times as many developers as Codex. That ratio subsequently fell to approximately 2.4.

But Anthropic's lead in absolute adoption increased from 15 to 23 percentage points. Codex is growing faster relative to its smaller base, while Claude Code is still reaching more new developers.

That's the catch with the OpenAI comeback story. Codex is catching up by one measure and falling further behind by another. Both are true.

Metric Claude Code OpenAI Codex
Workplace adoption, January 2026 18% 3%
Workplace adoption, May–July 2026 39% 16%
Adoption increase +21 points +13 points
Relative adoption growth 2.2× 5.3×
Developer awareness, latest JetBrains survey 94% 65%
Stack Overflow 2026, past-year usage 65.5% 29.5%
Workplace adoption relative to competitor 2.4× Codex 41% of Claude Code
Latest workplace adoption ranking First Third
Who is really ahead in frontier AI, OpenAI or Anthropic? Anthropic, on 3 lines of 5: $17B of third-quarter sales against about $10B, and the top model on the experts' index. OpenAI keeps the crowd and has raised more money. Chart from NewMarketPitch's Frontier AI Labs report.
Anthropic Today reportThis page and 70+ morePDF Get the report

Just a preview. If a number here looks old, no worries: your report comes with the latest data.

Has Codex become a major coding tool already?

OpenAI Codex has already reached several million weekly users, making its comeback much more substantial than a temporary burst of developer interest.

In June 2026, OpenAI reported more than 5 million weekly active Codex users, more than six times the level recorded around the February launch of its desktop application. Five million weekly users arriving this fast points to an opening, and we look at where the biggest opportunities are in our frontier AI labs market report.

That implies an earlier baseline below approximately 833,000 weekly users, although OpenAI did not disclose the exact figure. Reaching more than five million within roughly four months represents exceptionally fast growth for a relatively specialized productivity tool.

However, Codex's user base extends beyond programming. According to OpenAI, approximately 20% of its users were knowledge workers using the product for reports, spreadsheets, presentations, research, and automation.

Applying that percentage to the disclosed total gives a rough figure of four million weekly users outside the knowledge-worker category, primarily developers. This is our calculation, rather than a separately reported developer count, and the underlying total was already above five million.

Some Codex growth therefore comes from people who never used Claude Code as their main programming environment. Still, OpenAI has demonstrated that Codex can attract millions of regular users, which was far from obvious when Anthropic first established its lead.

Is Claude Code still developers' favorite AI coding agent?

Claude Code remains the most widely used dedicated coding agent among professional developers, and a remarkable number of its users consider it their main tool.

In JetBrains' latest research, 31% of professional developers identified Claude Code as their most-used AI coding tool, compared with 39% who reported using it at work. That gives a ratio of approximately 79% between primary-tool preference and overall workplace adoption.

It's an unusually strong result for a product competing against alternatives that developers can often try for little additional effort.

Stack Overflow's 2026 Developer Survey reached a similar conclusion using a different method. Among 12,255 respondents answering its coding-agent question, 65.5% had used Claude Code during the preceding year, compared with 58.7% for GitHub Copilot, 29.5% for Codex, and 25.5% for Cursor.

The two surveys measure different things. JetBrains tracks workplace adoption among professional developers, while Stack Overflow asks respondents which tools they have used over the past year. Their percentages therefore cannot be compared directly.

But both surveys put Claude Code ahead. And that's a lot more convincing than judging the market from whichever model is generating the most excitement on social media this week.

Are OpenAI's coding models finally better than Claude?

OpenAI has caught up with Anthropic on several demanding coding benchmarks, and GPT-6 Astra now beats leading Claude models in some important tests.

OpenAI reported a 57.9% score for GPT-6 Astra on Terminal-Bench 4.0, compared with 55.8% for Claude Fable 5.1 and 52.6% for Claude Opus 5. Its previous GPT-5.6 Sol model scored just 37.3%.

The improvement from Sol to Astra was 20.6 percentage points, far larger than Astra's narrow advantage over Claude Fable 5.1.

On DeepSWE v1.1, Astra reached 74.1%, narrowly ahead of Claude Opus 5 at 73.7%. The results were reversed on the Artificial Analysis Coding Agent Index v1.4, where Claude Opus 5 scored 68.1 and Astra reached 67.0.

OpenAI's published FrontierCode results also show Claude Fable 5 narrowly ahead of Astra on both the Main and Extended evaluations.

These figures come from OpenAI's model release evaluations, so the testing configurations and agent environments deserve consideration. Independent testing across identical workflows remains important.

Nevertheless, the performance gap that previously made Claude an obvious recommendation for many programmers has largely disappeared. OpenAI now has a credible case for being the better choice on specific coding tasks, particularly terminal-based automation and some repository-level engineering challenges.

OpenAI or Anthropic: who is closer to making a profit? Anthropic, by its own figures: an $81B sales pace and a $2.5B adjusted profit last quarter. OpenAI sells about $50B a year and forecasts about $280B of cash burn to 2030. Chart from NewMarketPitch's Frontier AI Labs report.
Anthropic Today reportThis page and 70+ morePDF Get the report

This is a preview page. Spotted an outdated number? That’s normal here. The report you get is always the freshest version.

Do better coding benchmarks actually make developers switch tools?

Better coding performance is the biggest reason developers change AI tools, which gives OpenAI's recent model improvements a real opportunity to win users.

Stack Overflow asked 13,144 respondents in its 2026 Developer Survey whether they had switched AI tools for work during the preceding year. Better output quality was the most common reason, cited by 25.1% of all respondents. Output quality beating cost as a reason to switch is why we check who leads on benchmarks and who only claims it in our frontier AI labs report.

Another 10.6% switched because their organization required or recommended a different tool, while 7.4% cited better workflow integration and 6.4% lower cost.

These percentages refer to all respondents, including those who did not switch. Among developers who actually changed tools, quality accounts for an even larger share.

The results help explain why coding benchmarks attract so much attention. If an agent fixes a difficult bug correctly, understands a large repository, and avoids breaking existing functionality, developers are willing to reconsider their current setup.

But benchmark wins don't always translate into better daily performance. A model that excels at solving isolated tasks may still struggle with a long debugging session, unfamiliar project conventions, or changes spanning dozens of files.

OpenAI's Astra introduced improved context preservation in Codex, allowing agents to retrieve information from earlier context windows instead of depending entirely on compressed summaries. That directly addresses a frustrating problem in long coding sessions: an agent forgetting why an earlier fix failed.

For developers deciding between Codex and Claude Code, reliability over several hours of work may matter more than a small difference in headline benchmark scores.

Is Codex winning the developers who use AI for almost everything?

Codex has attracted a particularly AI-heavy group of developers, including people who already delegate most of their coding work to agents.

In an August 2026 analysis, JetBrains found that 42% of Codex users reported generating more than 80% of their code with agents. Among developers who identified Claude Code as their most-used AI coding tool, the corresponding figure was approximately 32%.

The groups aren't defined in exactly the same way, so the ten-point difference should be interpreted carefully. Still, the comparison suggests that Codex appeals strongly to developers comfortable handing over substantial implementation work.

JetBrains also reported that around 37% of Codex users said they did not write code without AI assistance at all.

These are the people already testing how far autonomous programming can go. They care about parallel execution, large changes, context handling, and how much human supervision an agent needs.

In the broader JetBrains sample, approximately 31% of developers primarily used agents, while 47% combined AI assistance with manual coding.

Codex's success among intensive users could help OpenAI influence how agent-first development evolves. We don't yet know whether that audience translates into higher customer retention or more developers replacing Claude Code.

Is Anthropic still adding developers faster than OpenAI?

Anthropic is currently gaining more developer adoption in absolute terms, even as Codex grows much faster relative to its starting point.

The size of Claude Code's user base has not stopped its expansion. JetBrains recorded a 21-point adoption increase during the first part of 2026, compared with Codex's 13-point gain. Anthropic therefore added roughly 62% more adoption points over that measured interval.

Anthropic's business disclosures tell a similar story. During its February 2026 Series G announcement, the company reported that Claude Code had exceeded $2.5 billion in annualized revenue run rate, more than doubling since the beginning of the year.

It also said weekly active Claude Code users had doubled since January, business subscriptions had quadrupled, and enterprise customers generated more than half of the product's revenue.

A separate analysis cited in Anthropic's announcement estimated that Claude Code was responsible for roughly 4% of public GitHub commits worldwide, twice the estimated share from the preceding month. That estimate should be treated as an outside measurement rather than an audited GitHub statistic.

Anthropic's coding business was still expanding rapidly while OpenAI was making its comeback. The idea that Codex's revival has already come at Claude Code's expense simply doesn't fit these numbers.

Which AI lab's sales grow the fastest? Anthropic, among labs that report their sales: its third-quarter sales were 12 times a year earlier. Mistral's pace grew faster, but from about $20M. Chart from NewMarketPitch's Frontier AI Labs report.
Anthropic Today reportThis page and 70+ morePDF Get the report

Preview only. Some numbers may have changed since we made this page, but your report will be up to date.

Does Claude Code make more money than OpenAI Codex?

Anthropic has established a multibillion-dollar annualized business around Claude Code, but we still don't have enough comparable revenue data to say whether it makes more money than Codex.

Anthropic's disclosed run rate of more than $2.5 billion is particularly impressive because Claude Code was introduced publicly only in 2025. A run rate annualizes revenue at a recent pace, rather than measuring money earned over a completed financial year. A coding product at a $2.5 billion run rate so soon after launch is one business model at work; we compare which are winning in our report on frontier AI labs.

Enterprise customers contributed more than half of Claude Code revenue, showing that Anthropic was making substantial money beyond individual subscriptions.

OpenAI has published large Codex usage figures, but no directly comparable standalone revenue figure in its June disclosure.

The comparison is complicated by product packaging. Codex can be accessed through ChatGPT subscriptions, business offerings, and APIs, making it difficult to isolate exactly how much revenue comes from programming.

There is also a difference between the customer populations. Claude Code is strongly associated with software development, whereas OpenAI is expanding Codex into general knowledge work.

Anthropic currently has the clearer evidence of coding-specific monetization. OpenAI could have a larger commercial opportunity through its broader customer base, but the revenue needed to demonstrate that advantage hasn't been disclosed.

Could cheaper OpenAI models convince developers to leave Claude?

OpenAI's lower costs could persuade developers to use Codex for more work, especially when they run agents repeatedly throughout the day.

OpenAI's GPT-6 Astra release offers a concrete example. On Terminal-Bench 4.0, Astra scored 57.9%, compared with Claude Fable 5.1's 55.8%, while OpenAI estimated that Astra cost approximately 63% less per task.

That suggests a team could achieve similar or slightly better performance at substantially lower inference cost under those benchmark conditions.

For occasional coding assistance, a price difference may barely affect the purchasing decision. For teams running agents through hundreds of automated tasks, especially when each task involves repeated testing and revisions, the economics become much more important.

The savings won't necessarily match benchmark estimates in production. Real costs depend on task complexity, context length, token consumption, caching, retries, and the pricing terms available to customers.

Lower inference costs also don't automatically translate into cheaper subscriptions. Developers who use fixed-price plans may care more about usage limits and completed work than the underlying API bill.

Stack Overflow's switching survey places output quality well ahead of cost. OpenAI's pricing advantage becomes most compelling when developers are satisfied that Codex can complete the same work as Claude Code.

Is Codex growing because OpenAI makes it easier to try?

OpenAI has made Codex much more accessible, and the jump in developer awareness suggests that its distribution strategy is working.

JetBrains found that awareness of Codex increased from 27% of professional developers in January 2026 to 65% by May–July. That is a 38-point increase within roughly half a year.

Awareness alone doesn't guarantee usage, of course. But Codex adoption grew even faster proportionally over the same period.

If we divide adoption by awareness, the ratio increased from about 11% to 25%. This isn't a true conversion rate because the surveys don't track the same individuals, but it shows that more developers were using Codex relative to the number who knew about it.

Product availability likely helped. OpenAI expanded Codex through its desktop application, ChatGPT subscriptions, APIs, and development environments.

JetBrains also added Codex support alongside Claude Agent, GitHub Copilot, and OpenCode in its IDE ecosystem, allowing developers to experiment without giving up their existing editor.

More recently, a Stackness analysis of third-quarter developer-tool activity found that Codex increased its VS Code Marketplace installs by 35.8% during the quarter. Claude Code grew 38.4% on the same measure.

Install counts include repeat installations and should not be confused with active users. Still, this provides a separate indication that both tools continue expanding beyond their existing audiences.

Who is fighting whom in frontier AI? OpenAI leads Google on users, Anthropic leads OpenAI on sales, and Alibaba leads DeepSeek on downloads. Alibaba and Kling split video, SSI and Thinking Machines split money and product. Chart from NewMarketPitch's Frontier AI Labs report.
Anthropic Today reportThis page and 70+ morePDF Get the report

This is just a preview page. Don’t worry if a number looks old: the report gets refreshed, and you’ll always get the latest version.

Are developers leaving Claude Code for Codex or using both?

Developers are increasingly comfortable using multiple coding agents, and there's currently no reliable public figure showing a mass migration specifically from Claude Code to Codex.

The 2026 Stack Overflow survey illustrates how widespread overlapping usage has become. Respondents could report using several coding agents during the preceding year, and the combined percentages for Claude Code, Copilot, Codex, and Cursor reached 179.2%.

Those percentages cannot represent exclusive market shares because many respondents selected more than one tool.

The same survey found that roughly two-thirds of respondents had switched AI tools for work during the year. However, the question didn't identify which providers developers left and which they joined.

A programmer might use Claude Code for complex refactoring, Codex for automated implementation, and Copilot inside an established IDE. Another might test a new model for a week before returning to a familiar tool.

The growing availability of multiple agents inside development environments makes this behavior easier.

OpenAI can gain users without taking existing users away from Anthropic, and Anthropic can keep growing while individual programmers increasingly experiment with Codex.

To establish a genuine reversal, we'd need data tracking which developers changed their primary coding agent from Claude Code to Codex and continued using it over several months.

Is GitHub Copilot losing more ground than Claude Code?

GitHub Copilot has lost more visible ground than Claude Code, suggesting that the coding-agent boom is putting greater pressure on older forms of AI coding assistance.

JetBrains found that Copilot's workplace adoption fell from 29% in 2025 to 21% in its latest survey. Cursor also declined from 18% in January 2026 to 12% during May–July, despite becoming more widely recognized among developers. Copilot and Cursor slipping while others grow is why we weigh who is ahead, and who is falling behind, in our report on frontier AI labs.

Meanwhile, Claude Code and Codex were both growing.

The changing rankings suggest that developers are increasingly willing to use agents that can handle complete tasks, rather than relying exclusively on tools built around suggestions and assistance inside an editor.

However, Copilot and Cursor have also introduced more advanced agent capabilities. The products are evolving, so the distinction between an editor assistant and a coding agent is becoming less clear.

The third-quarter marketplace data adds some nuance. According to Stackness, GitHub Copilot Chat installs still increased by 3.2%, even though its survey adoption had declined. An installed extension can remain in place while developers use it less frequently.

Coding tool Earlier workplace adoption Latest adoption Change
Claude Code 3% (spring 2025) 39% +36 points
GitHub Copilot 29% (spring 2025) 21% −8 points
OpenAI Codex 3% (January 2026) 16% +13 points
Cursor 18% (January 2026) 12% −6 points
Google Antigravity 6% (January 2026) 6% Unchanged
OpenCode No comparable earlier figure 7% Not established
JetBrains AI No comparable earlier figure 9% Not established

The figures come from JetBrains' developer research. Earlier baselines vary by product, and developers can use several tools simultaneously.

Who has held the top AI model, month by month? Anthropic and Google, taking turns: over the last 12 months Anthropic topped the LMArena users' vote for 7 months and Google for 5. OpenAI never did. Chart from NewMarketPitch's Frontier AI Labs report.
Anthropic Today reportThis page and 70+ morePDF Get the report

Just a preview. If a number here looks old, no worries: your report comes with the latest data.

Can OpenAI's huge ChatGPT audience help Codex beat Anthropic?

OpenAI has a much larger general-purpose AI audience to introduce to Codex, giving it a distribution advantage that Anthropic cannot easily match.

In September 2026, OpenAI reported that its products served more than one billion weekly active users and 2.5 million businesses.

That reach creates a large pool of potential Codex customers already familiar with OpenAI's interfaces, models, and subscriptions.

The company is also pushing Codex beyond traditional software development. In June, OpenAI reported that knowledge workers represented approximately one-fifth of Codex users and were adopting the product more than three times as fast as developers.

These users were applying Codex to research, data analysis, document creation, spreadsheet automation, and lightweight applications.

As programming agents become simpler to operate, more people can use them without being professional programmers.

Anthropic is pursuing a related opportunity through Claude and its workplace products, but Claude Code has built a particularly strong identity among developers.

OpenAI's enormous audience could eventually give Codex the larger overall user base. Whether it becomes the preferred tool for professional programming is a separate question, and one where Anthropic remains ahead.

Could Anthropic's enterprise customers make Claude Code harder to replace?

Anthropic's growing enterprise business makes Claude Code harder to displace, particularly inside organizations that have already integrated it into their development processes.

In its February 2026 funding announcement, Anthropic said more than 500 customers were spending over $1 million annually across its products, compared with approximately a dozen two years earlier.

That larger customer figure covers Anthropic's overall business, including products beyond Claude Code. Still, the rapid increase shows how deeply the company's models were spreading into corporate workflows.

For coding agents, enterprise adoption involves more than buying individual subscriptions. Companies need to manage repository access, permissions, security policies, spending limits, and the way agents interact with internal systems.

Once those processes are established, changing providers requires additional testing and approval, even if a competing model performs better on a benchmark.

OpenAI also has extensive enterprise relationships, including businesses already using ChatGPT and OpenAI APIs. Codex can benefit from that existing presence.

The unresolved question is which provider is winning the largest new deployments and renewals. Neither company has released enough directly comparable coding-agent contract data to settle it.

For now, Anthropic's strong enterprise revenue gives Claude Code a firmer position than its popularity among individual developers alone would suggest.

Is OpenAI improving coding AI faster than Anthropic?

OpenAI has made an impressive technical comeback, but Anthropic is releasing strong coding models frequently enough that a lasting OpenAI lead remains uncertain.

Anthropic introduced Claude Opus 4.7 in April 2026, Opus 4.8 in May, Claude Sonnet 5 in June, Opus 5 in July, and Opus 5.5 in September. Its Claude Fable models have also pushed performance higher on difficult agent tasks.

OpenAI's newer GPT models have caught up across several evaluations while emphasizing lower costs and more reliable long-running workflows.

Both companies are also improving the surrounding agents, which can matter as much as the model itself.

For example, Codex's experimental context-preservation system lets Astra retain notes and search earlier context windows. This can help with large refactors where an agent needs to remember failed attempts, previous test results, and requirements from much earlier in the session.

Anthropic has similarly focused on longer-running agent workflows, stronger tool use, and improved performance on substantial coding tasks.

The release pace makes long-term technical predictions difficult. A model can establish a narrow lead only to lose it a few weeks later.

Developers now have several genuinely competitive options for advanced AI coding. OpenAI is firmly back in that group.

Is Anthropic catching up with OpenAI in fundraising? Fast. Anthropic raised about $95B in 2026 alone and has now raised about two thirds of what OpenAI has, up from about half at the end of 2025. Chart from NewMarketPitch's Frontier AI Labs report.
Anthropic Today reportThis page and 70+ morePDF Get the report

This is a preview page. Spotted an outdated number? That’s normal here. The report you get is always the freshest version.

What would prove that OpenAI has won developers back from Anthropic?

OpenAI would need to show that more developers are choosing Codex as their main coding agent and keeping it that way, especially after previously relying on Claude Code.

The most useful evidence would come from surveys tracking individual developers across multiple quarters. Those could show how many users moved from Claude Code to Codex, how many moved in the other direction, and how many simply added another tool. Tracking who really switches coding tools is hard; we look at the biggest unsolved problems in the market in our report covering frontier AI labs.

Primary-tool preference would be especially revealing. It's relatively easy for a developer to install Codex and experiment with a new model. Making Codex the tool used for most daily programming is a much stronger commitment.

Retention also matters. New model releases create bursts of interest, but developers tend to stick with tools that consistently understand their repositories, complete difficult tasks, and fit their workflow.

Commercial figures could help settle the question too. Comparable coding-specific revenue, enterprise subscriptions, and renewal rates would reveal whether developer enthusiasm is turning into a durable business advantage.

For now, the available public data establishes that Codex is growing rapidly. It doesn't yet show a sustained wave of developers making the switch from Claude Code.

So, is OpenAI winning coders back from Anthropic?

Partly yes. OpenAI has regained serious credibility among developers and turned Codex into one of the fastest-growing AI coding tools, but Anthropic remains the leader in professional adoption.

Codex now has millions of weekly users, much broader developer awareness, and coding models that can match or outperform Claude on several demanding evaluations. The product has become a credible choice for developers who want to delegate substantial engineering work to agents.

Anthropic, however, has continued growing while OpenAI catches up. Claude Code remains the most-used dedicated coding agent in major developer surveys, and its enterprise business has already reached substantial commercial scale.

As seen above, the most revealing comparison is that Codex's relative adoption gap narrowed while Claude Code's absolute lead widened. Both companies gained ground, but Anthropic was still adding more developer adoption points over the measured period.

We also found no reliable public evidence quantifying how many developers have abandoned Claude Code specifically to return to OpenAI.

Our judgment is that OpenAI has won back developers' attention and their willingness to experiment, which is already a major turnaround. Winning back their primary coding workflow is the harder challenge, and Anthropic currently has the stronger position.

The competition is much closer than it was when Claude Code first took off. But until Codex starts overtaking Claude Code in sustained daily preference and enterprise adoption, calling this an OpenAI victory would be premature.

ChatGPT, Gemini, Meta AI, Claude: who has the most users? ChatGPT, with 1.2 billion people a week. Gemini and Meta AI each claim about a billion a month, and Anthropic gives no count for Claude. Chart from NewMarketPitch's Frontier AI Labs report.
Anthropic Today reportThis page and 70+ morePDF Get the report

Preview only. Some numbers may have changed since we made this page, but your report will be up to date.

Our methodology

This analysis examines whether OpenAI is winning software developers back from Anthropic by comparing developer adoption, primary-tool preferences, coding-agent performance, switching behavior, product accessibility, and commercial traction. We assessed these dimensions separately before bringing the evidence together to judge how far OpenAI's comeback has progressed.

Our main adoption evidence comes from JetBrains' research on professional developers and Stack Overflow's 2026 Developer Survey. We use JetBrains to compare workplace adoption, changes over time, developer awareness, and primary-tool preference. Stack Overflow provides a separate view of past-year coding-agent usage and the reasons developers change AI tools. The surveys use different populations and questions, so their percentages are interpreted within each survey rather than combined into a single market-share estimate.

We distinguish relative growth from absolute adoption gains. We also separate increased usage from actual competitive displacement: developers may adopt Codex alongside Claude Code without changing their primary coding tool. Where the surveys do not track individual developers moving between providers, we avoid treating new adoption as evidence of direct switching.

For technical performance, we examined OpenAI's published GPT-6 Astra evaluations, including Terminal-Bench 4.0, DeepSWE v1.1, and FrontierCode, alongside the independent Artificial Analysis Coding Agent Index. We compare benchmark results within their stated evaluation conditions and distinguish published per-task cost estimates from real-world subscription prices or production costs.

Commercial traction is assessed using OpenAI's Codex user disclosures and Anthropic's funding announcement, including Claude Code revenue run rate, enterprise participation, and reported user growth. We keep weekly active users, annualized revenue, workplace adoption, and enterprise customer counts separate because they measure different aspects of the business. We also examined product announcements and Stackness's VS Code Marketplace analysis as supporting evidence for distribution and developer-tool activity.

Key developer research sources include JetBrains on AI coding-agent adoption trends, JetBrains on workplace AI tool usage, JetBrains on agent-generated code, Stack Overflow's coding-agent usage data, and Stack Overflow's AI tool-switching research.

OpenAI sources include the GPT-6 Astra release and evaluations, Codex adoption and knowledge-work expansion, the Codex desktop application announcement, Codex IDE and workflow upgrades, Codex for different professional workflows, and OpenAI's broader product adoption disclosures.

Anthropic sources include its Series G funding announcement and Claude Code commercial figures, Claude Opus model information, and the Claude Opus 5.5 release. Independent technical and ecosystem references include Artificial Analysis's coding-agent benchmarking methodology and Stackness's third-quarter developer-tools analysis.

We prioritize original survey findings, official disclosures, documented technical evaluations, and published ecosystem measurements. Company-reported results are treated as first-hand evidence of what each provider disclosed, while independent surveys and benchmark research provide separate checks on competitive positioning. Our final assessment gives the greatest weight to sustained workplace adoption and primary-tool preference, rather than temporary interest around new releases.