Luna Outcodes Opus, Fable Borrowed Time: OpenAI's $1 Model Just Made Anthropic's Flagship Look Expensive

OpenAI shipped GPT-5.6 as three separate large language models—Sol, Terra, and Luna—each with distinct training, pricing, and capability ceilings, marking the first time the company has released a flagship family without built-in thinking-dial controls. Pricing for the lineup sits at $5 per million input tokens and $30 output for Sol, with Luna, the cheapest tier, priced at $1 input and $6 output. Claude Fable 5, Anthropic's most capable publicly available model, is listed at $10 per million input tokens and $50 output—double Sol's rate.

On the Artificial Analysis Coding Agent Index, Sol scored 80 against Fable's 77.2, completing the run in under half the time, with roughly half the tokens, at about a third of the cost. On Agents' Last Exam, which runs professional workflows across 55 fields, Sol hit 53.6% against Fable's 40.5%. In Terminal-Bench 2.1, Sol in ultra mode, running four subagents in parallel, posted 91.9% versus Fable's 83.1%. On the broader Intelligence Index, which aggregates nine benchmarks, Fable 5 leads GPT-5.6 by one point.

Luna, the $1/$6 tier, already outranks Anthropic's Opus 4.8 on coding benchmarks, putting pressure on Anthropic's subscription stack if Fable 5 exits the lineup on July 19. The U.S. government banned Fable 5 on June 12 after Amazon researchers demonstrated a jailbreak that turned the model into an unintended vulnerability scanner. Anthropic pulled the model globally for 19 days, rebuilt its safety classifier, and reinstated access on July 1 with a compressed usage window. The company has since pushed back its planned move to a usage-credits paywall three times—from July 7 to July 12 to July 19—issuing each extension hours before the cutoff rather than via formal blog post. "We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code's weekly rate limits 50% higher, through July 19," Anthropic's @claudeai account posted on July 12, 2026.

The repeated extensions come as Fable's technical standing has narrowed against OpenAI's mid-range offerings, with the model running under tightened weekly rate limits even on paid subscriptions. On creative-writing prompts outside standard coding benchmarks, including a request to send a character named Jose Lanz back from 2150 to the year 1000, induce a time-travel paradox, and conceal the cause from him until his return, both Sol and Fable returned outputs closer to novelettes than short stories, and both failed to fully honor the prompt's core constraint. Fable 5's $10/$50 token pricing and the looming July 19 deadline have turned its benchmark parity with GPT-5.6 into a subscription-tier liability for Anthropic, with Opus 4.8 the default fallback once Fable access ends.