Reading
Replit Made ChatGPT Its Default. That's a Deal, Not a Verdict
Replit's CEO says OpenAI now beats Claude on cost and on Replit's own benchmarks. Here's what he actually said, what's disputed, and why that proves less than it sounds.
Prices last read 2026-10-31
Short answer
No. Replit's switch doesn't prove ChatGPT is now the better coding model. It's a business decision that also mentions performance.According to TechCrunch, Replit CEO Amjad Masad said the decision came down to 'cost, partnership, and performance' and pointed to Replit's own benchmarks. Claude hasn't gone anywhere: paying Replit users can still pick it as the Agent's primary model. What changed is the default, and TechCrunch notes that most customers never change the default.Source: Replit makes OpenAI its default over Anthropic, CEO says (TechCrunch) · 2026-10-31; Model selector (Replit Docs) · 2026-10-31
What Masad actually said, and how quickly it softened
TechCrunch reports that on stage at Disrupt, Masad said Replit was no longer working with Anthropic and added, 'I think ChatGPT is better now.' He said OpenAI's Sol and Astra models 'consistently' do better at building vibe-coded apps from scratch. Then the message got walked back. Over the weekend he told TechCrunch that 'Replit remains model agnostic,' that users can still choose Claude, and that OpenAI is now the 'preferred partner for powering Replit Agent' based on 'performance, economics, and the strength of our partnership.' TechCrunch also changed its headline from 'no longer working with Anthropic' to 'makes OpenAI its default.'
Source: Replit makes OpenAI its default over Anthropic, CEO says (TechCrunch) · 2026-10-31
The status of the relationship is still disputed. A person familiar with the partnership told TechCrunch it is 'still in place.' TechCrunch also points out that Replit announced its OpenAI partnership in August, and that Replit's own blog was promoting Claude models running in production only weeks before Disrupt. IBTimes Singapore, working from the same reporting, adds that Replit hasn't published comparable cost figures, so nobody outside the company can check how big the price advantage is.
Source: Replit makes OpenAI its default over Anthropic, CEO says (TechCrunch) · 2026-10-31; AI Coding Platform Dumps Anthropic for OpenAI, Says ChatGPT Models Cheaper and Better (IBTimes Singapore) · 2026-10-31; Replit introduces Free Mode (Replit Blog) · 2026-10-31
Replit's own published numbers tell a messier story
Here's the detail that got lost in the stage drama. Replit's engineering post from September 29 credits both GPT-6 Astra and Claude Fable 5.1 with better code quality, good enough that Replit could run its code-review subagent less often without its eval scores dropping. On Replit's published DeepSWE and Terminal-Bench runs, Astra at its highest effort setting beat Replit Agent's Max mode, but at more than twice the cost per task. At low effort Astra was cheaper but scored lower. That post doesn't mention Sol at all, and TechCrunch didn't link the public benchmark Masad referred to.
Source: Free the models (Replit Blog, Sept 29, 2026) · 2026-10-31; Replit makes OpenAI its default over Anthropic, CEO says (TechCrunch) · 2026-10-31
My take: this is a pricing-and-partnership move dressed up as a quality verdict. If OpenAI were simply better at coding, Replit wouldn't need 'partnership' and 'economics' in the same sentence. The most overrated thing here is the soundbite 'ChatGPT is better now.' The most underrated thing is Replit's own September post, which treats Astra and Fable 5.1 as roughly equal on code quality and shows that Astra's best score costs a lot more. When a company's marketing and its engineering blog disagree, believe the engineers.
Who this actually affects
If you use Replit Agent on Core or Pro, you can still choose the primary model yourself in Power or Max Mode. Power Mode lists GPT-6 Sol alongside Claude Sonnet 5, and Max Mode lists GPT-6 Astra alongside Claude Fable 5.1 and Claude Opus 5.5. Free Mode users can't choose: they get automatic routing, shown as 'Auto.' If you pay for ChatGPT or Claude directly, nothing about your subscription changed because of this.
Source: Model selector (Replit Docs) · 2026-10-31
- Open Agent ModeIn Replit's prompt box, select Agent Mode.
- Pick Power or MaxModel choice is available only in these modes, on Core and Pro plans.
- Change the Primary modelOpen the Primary model menu and choose a model. Your choice applies from your next message. If you don't see Primary model, the selector isn't available on your account yet.
Source: Model selector (Replit Docs) · 2026-10-31
Code generated or suggested by our AI systems may be erroneous or incomplete.
Replit Terms of Service, Section 6a (last updated August 3, 2026)Source: Replit Terms of Service · 2026-10-31
Why this matters: whichever model sits behind the default, Replit's terms leave checking the generated code to you. The same terms also let Replit modify or remove features 'at any time and for any reason,' so your default could change again without warning.
Source: Replit Terms of Service · 2026-10-31
Before you switch subscriptions over this headline
- Run the same real task from your own repo on both Astra and Fable 5.1 in Replit's selector, and compare the results.
- Track what each run cost you, not what a CEO said on stage.
- Don't treat vendor-run benchmarks as neutral, whoever publishes them.
- Remember that the default reflects a partnership. It isn't a ranking.
Bottom line
Keep the Claude or ChatGPT plan that already works for how you code. Replit picked a partner, not a winner, and its own engineering blog says so more honestly than its CEO did.
Optional: compare third-party seller listings for ChatGPT plans (a reseller listing, not official OpenAI access or an endorsement). That link carries our partner code Sk52g, which is how a purchase is attributed to this site. Promotion code UPD3S.
Sources (all read 2026-10-31)