OpenAI’s GPT-5.6 raises pricing pressure on Anthropic’s Fable 5

OpenAI’s GPT-5.6 rollout gives AI users a more segmented menu, and it puts fresh price pressure on Anthropic’s Claude Fable 5. For retail investors watching the private AI race from the sidelines, the key issue is straightforward: cheaper, capable models can change which companies developers choose to build on.

According to Decrypt, OpenAI is not releasing GPT-5.6 as one large language model with adjustable “thinking” settings. A large language model, or LLM, is the AI system that generates text, code, and other outputs after being trained on large datasets. GPT-5.6 instead arrives as three separate models: Sol, Terra, and Luna.

Decrypt reported that the three OpenAI models differ in training, cost, and performance limits. The main comparison is between Sol and Claude Fable 5, which Decrypt described as Anthropic’s strongest public model currently available.

Pricing is the first pressure point

For developers, model pricing is usually measured in tokens, the chunks of text an AI system reads and produces. OpenAI’s Sol costs $5 per million input tokens and $30 per million output tokens, according to Decrypt. Input tokens are the text sent into the model, while output tokens are the text the model generates back.

Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, Decrypt reported. That makes Fable 5 more expensive on both sides of the API bill.

Decrypt also reported that Sol is now ahead of Fable 5 on several developer-focused benchmarks. Benchmarks are standardized tests used to compare model performance, though they do not capture every real-world use case. The report added that Luna, OpenAI’s lowest-cost GPT-5.6 model at $1 per million input tokens and $6 per million output tokens, outranks Anthropic’s Opus 4.8 on coding.

Anthropic is managing access after a safety issue

Claude Fable 5’s recent history has been complicated. Decrypt reported that the U.S. government banned the model on June 12 after Amazon researchers identified a jailbreak that made the system operate as an unintended vulnerability scanner. A jailbreak is a prompt or method that gets an AI model to bypass its safeguards.

Anthropic then withdrew Fable 5 worldwide for 19 days, Decrypt reported. The company built a new safety classifier and restored the model on July 1 with a shorter access period.

Since then, Anthropic has extended access more than once. Decrypt reported that the company first planned to place Fable 5 behind a usage-credits paywall on July 7, then moved that date to July 12, and later pushed it again to July 19.

In a July 12 post on X, the Claude account said Anthropic was extending Claude Fable 5 access across paid plans through July 19. The same post said Claude Code’s weekly rate limits would remain 50% higher through that date.

The IPO race is still speculative

The AI competition is also showing up in prediction markets. A sponsored Myriad market listed Anthropic at 81% and OpenAI at 19% on the question of which company will hold an initial public offering first, with $14.3K in volume shown. Myriad says its odds are set by users buying and selling predictions, with Points and USD₮ markets available depending on country.

Those market odds are not a company filing or a confirmed timeline. They do show how closely traders and AI users are watching the same question: which private AI lab can turn model performance, pricing, safety controls, and developer demand into the stronger long-term business.

This story draws on original reporting from Decrypt.