Two Labs Drop Flagships In The Same Breath, Race To The Cheapest Million Tokens

Anthropic released Claude Opus 5.5 on Tuesday, and OpenAI answered within minutes by launching GPT-6 Sol and GPT-6 Luna, setting off a coordinated price cut between the two largest AI model providers. Anthropic said Opus 5.5 performs at the level of Claude Fable 5.1, its top-priced flagship, while costing 20% less than Opus 5 and 60% less than Fable 5.1. The new model arrived less than a month after OpenAI introduced GPT-6 Astra on September 3, which OpenAI described as "the most intelligent and aligned model in the world." Astra remains OpenAI's top-tier option, with Sol and Luna positioned as faster, lower-cost alternatives.

Pricing details published by each company show both vendors cutting token rates across the board. OpenAI priced GPT-6 Sol at $2 per million input tokens and $10 per million output tokens, down from $4 and $20, and set GPT-6 Luna at $0.10 and $0.50, down from $0.20 and $1.20. Anthropic listed Opus 5.5 at $4 per million input tokens and $20 per million output tokens, with cached reads at $0.20 per million tokens, which the company called a 60% reduction from Opus 5. Anthropic also said output generation has accelerated by more than 30%. OpenAI said caching and inference have been made more efficient on its new models.

The two vendors posted competing benchmark numbers. On Terminal-Bench 4.0, which scores agents on completing complex professional tasks inside a command-line interface, Anthropic reported Opus 5.5 at 66.4%, ahead of Fable 5.1's 55.8% and GPT-6 Astra's 57.9%. On FrontierCode, which evaluates whether an agent's code changes would be merged in a real engineering pipeline, Opus 5.5 scored 54.4% against Fable 5.1's 50.3%. On GDPval-AA v2.1, which uses Elo ratings across 44 occupations, Opus 5.5 scored 1846 against Fable 5.1's 1735 and Opus 5's 1708.

OpenAI, citing its own measurements, said GPT-6 Sol scored 33.2% on AutomationBench at its highest reasoning setting for $0.27 per task, compared with Claude Opus 5's 26.9% at more than 11 times the cost, and edged out Claude Fable 5.1 on the same test. On Agents' Last Exam, GPT-6 Sol at its highest effort scored 56.4%, which OpenAI said beats Claude Opus 5's best score at 60% lower cost per task. On OSWorld 2.0, GPT-6 Sol scored 60.5% against Claude Opus 5's medium-effort 60.3% at roughly 80% less cost. Anthropic noted that GPT-6 Astra's 41.4% on AutomationBench edged out Opus 5.5's 40.0%. GPT-6 Sol and Luna are live now in ChatGPT Work and Codex for Plus, Pro, Business and Enterprise tiers.

The release comes as Anthropic's CEO Dario Amodei has pushed for slower capability gains. "We must slow the pace at which we improve the capabilities of AI models," Amodei wrote earlier this month in an essay calling on the industry to let safety testing keep pace. CoinGape reported that Anthropic is preparing for a November IPO that could value the company near $2 trillion, though the company has not publicly confirmed those terms.