Perplexity adds Fable 5.2 to Computer and Teammates, claims lead over Gemini Argon

Perplexity CEO Aravind Srinivas said that Fable 5.2 is now available on Perplexity, where it can be used inside Computer or as the model behind a Teammate. He also said it beats Google’s Gemini Argon “by a solid margin” on Perplexity’s WANDR research benchmark.

The news came in a post on X on November 9. Srinivas wrote: “Fable 5.2 now available on Perplexity. Use it inside Computer, or create Teammates with Fable. Fable beats Gemini Argon by a solid margin on Wide and Deep Research tasks (WANDR).” The post gives no scores, pricing or plan details, and a fuller announcement was not available when this article was written.

Where Fable 5.2 shows up in Perplexity

According to the post, there are two entry points. The first is Computer, Perplexity’s agent product, which hands tasks to a mix of models. The second is Teammates, the personal work agents Perplexity introduced on November 5. Users can now pick Fable when they create one.

Teammates were rolled out to Max users on the web first, according to Srinivas’s launch post, with native apps to follow. Perplexity’s Max plan costs $200 a month, as covered in our pricing guide. The post does not say whether Fable 5.2 is limited to Max, whether Pro subscribers get it, or whether it costs extra credits or usage. Those points are unconfirmed.

The model comes from Anthropic, which launched Claude Fable 5.2 recently. The post does not describe any change to how Perplexity runs it.

A benchmark Perplexity built and runs itself

The comparison with Gemini Argon rests on WANDR, short for Wide and Deep Research. Perplexity open-sourced the benchmark in July, according to Crypto Briefing. It is aimed at agents doing broad searching combined with deeper investigation, and Perplexity presents it as an evolution of the earlier WideSearch framework.

No figures accompany the Fable claim. Without scores, the size of the “solid margin” cannot be checked, nor can the settings used, such as which agent harness ran each model. The claim comes from the company that designed the test and sells the product, so it should be treated as Perplexity’s own result until others reproduce it.

Strategy: Perplexity keeps selling the harness, not one model

The post fits a pattern in how Perplexity positions Computer. At its February launch, press coverage described it as a system that routes work across many models, with Google’s Gemini used for deep research and Anthropic’s Opus as a core reasoning engine, per AlternativeTo. Adding the newest model within days, and naming a rival lab’s flagship as the comparison, tells customers that the value is in the product layer around the models.

Demand for that layer looks strong, if Srinivas’s own numbers hold. In earlier posts he said Teammates usage has risen every day since launch and that the median Perplexity employee gets 7x more agent-hours per week from teammates than in September. Both are company claims. They do suggest why Perplexity would want its strongest research model available to Teammates quickly: longer-running agents consume more model capacity and expose quality differences faster.

The benchmark choice matters too. WANDR was built around the wide-and-deep research work that Perplexity’s search stack is designed for, so a lead on it speaks to that workload and not to coding or other agent tasks.

What readers can check next

Three things would settle the claim: published WANDR scores for Fable 5.2 and Gemini Argon, a list of which plans can select Fable, and any independent run of the benchmark. Srinivas also said more features would ship this week, so details may follow soon.