Anthropic's Fable 5 spent three days as the world's most capable AI model before the US government pulled it offline.
Back
Anthropic Fable 5 tops global AI ranking as China-US gap widens to 13%
Anthropic's Fable 5 and Mythos 5 topped the global model intelligence leaderboard, widening the China-US AI capability gap to 13% from 8%, Jefferies said — a disparity the broker expects to grow as US labs gain preferential access to Nvidia's next-generation chips.
"The disparity in computing resources stems from US AI laboratories having greater access to advanced AI chips, including Nvidia's Blackwell series," Jefferies analysts wrote in a report dated June 14. The broker said that while some Chinese models train overseas using Nvidia hardware, the number of advanced chips accessible to Chinese companies will remain far below that of US peers.
Fable 5, launched June 9, scored 80.3% on SWE-Bench Pro against GPT 5.5's 58.6% and 95% on SWE-Bench Verified, according to published benchmarks. The model offered a 1-million-token context window and 128,000 output tokens at $10 per million input tokens and $50 per million output — double GPT 5.5's pricing of $5 and $30. The US Commerce Department ordered Anthropic to suspend access on June 12, citing a jailbreak vulnerability, making GPT 5.5 the top model available to consumers.
The widening capability gap reinforces the US AI leadership narrative, supporting premium valuations for Nvidia and US AI infrastructure providers. But Jefferies flagged a bigger concern: API pricing in China continues to decline — MINIMAX-W cut prices 33% in June — even as hardware costs surge, threatening returns on AI investment across Chinese laboratories.
Computing Gap Drives Model Divergence
Jefferies attributed the widening intelligence gap to structural differences in chip access. US AI labs have greater access to Nvidia's Blackwell series, and the company is set to begin shipping its next-generation Rubin chips in the second half of 2026, potentially expanding the computing disparity further. Chinese companies are developing alternatives using Huawei's "Logic Folding" technology, but at potentially higher costs, the broker said.
The gap is measurable. Fable 5 led the Code Arena by 98 Elo points, scoring 1,665 to GPT 5.5's 1,501, and on FrontierCode Diamond — a benchmark for the most difficult programming tasks — Fable 5 scored 29.3% while GPT 5.5 managed 5.7%, according to published leaderboard data. Only on Terminal-Bench 2.0, which evaluates interactive terminal-based coding, did the gap narrow, with GPT 5.5 scoring 82.7% against Fable 5's roughly 88%.
Price War Collides With Rising Costs
Jefferies said its biggest concern is not the capability gap itself but the pricing dynamics in China. From March to June, average API pricing in the US rose 89%, partly driven by higher hardware costs. Anthropic doubled the API price of Fable 5 compared with Opus 4.8. In China, however, API pricing remained stable before falling 2% month over month in June after MINIMAX-W cut prices by 33%.
"To achieve an acceptable return on investment in AI, API pricing must rise while token consumption continues to surge," Jefferies analysts wrote. "Price competition among Chinese AI laboratories is currently our biggest concern." With memory costs rising sharply and Huawei's alternative chip technology carrying potentially higher production costs, hardware expenses will continue to climb, the broker added.
The US government's export control directive against Fable 5 — which also cited the model's "anti-distillation" capabilities — may make it more difficult for China's open-source models to catch up through knowledge transfer. Anthropic has disputed the severity of the finding, saying the vulnerabilities identified are minor and also achievable on GPT 5.5 without bypass techniques, according to company statements.
For investors, the divergence creates a bifurcated opportunity. Nvidia, trading at elevated multiples on sustained AI demand from US labs, benefits from the structural computing advantage. Chinese AI companies face margin compression from price wars amid rising hardware costs, making near-term profitability elusive. Whether Fable 5 returns to the market depends on Anthropic's negotiations with the government over the export control classification — a resolution that could reshape the competitive landscape again.
This article is for informational purposes only and does not constitute investment advice.
[1] Jefferies: Anthropic Fable/Mythos 5 Tops Global Model Intelligence Ranking, China-US AI Gap Widens to 13%[2] The Fable 5 AI Model Just Went Dark. What the US Government’s Playbook Means for AI Companies—and Every Business Leader[3] Fable 5 was beating GPT 5.5 on every major benchmark. Then the US government pulled it offline.[4] Claude Fable 5 vs GPT 5.5: Is this why the Trump admin banned one and not the other?
