Two Flagships Walk Into a Tuesday: OpenAI Cuts 50%, Anthropic Cuts 60% 🤖

OpenAI released two new models, GPT-6 Sol and GPT-6 Luna, on Tuesday, minutes after Anthropic launched Claude Opus 5.5, the first entry in its Claude 5.5 family. Both launches targeted the same problem — making flagship-grade AI cheaper — and both arrived within the same hour on September 22, 2026.

GPT-6 Sol and GPT-6 Luna sit below GPT-6 Astra, the model OpenAI called "the most intelligent and aligned model in the world" when it launched on September 3. Astra remains OpenAI's top pick for the hardest jobs; Sol and Luna are positioned as the faster, lower-cost options built for everyday work. OpenAI cut API pricing for both models by 50% compared with GPT-5.6's promotional rates per token. Sol now costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20. Luna dropped to $0.10 and $0.50, down from $0.20 and $1.20.

Anthropic's Claude Opus 5.5 was framed by the company as matching Claude Fable 5.1, its priciest flagship, on most tasks while costing 20% less to run than Opus 5 and 60% less than Fable 5.1. "We must slow the pace at which we improve the capabilities of AI models," CEO Dario Amodei wrote earlier this month in an essay calling on the industry to dial back capability gains. Opus 5.5 is the first model Anthropic has shipped since that essay was published.

OpenAI's performance case leans on AutomationBench, a Zapier-built benchmark that scores whether an AI agent can complete a full business workflow using 47 tools across sales, marketing, operations, support, finance and HR. GPT-6 Sol at its highest reasoning setting scored 33.2% for $0.27 per task; OpenAI said Claude Opus 5 at its top setting scored 26.9% at more than 11 times the cost per task, and that Sol also edged out Claude Fable 5.1 on the same test. On Agents' Last Exam, which grades long, economically valuable work across 55 sub-industries as a percentage completed, 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. For computer use, GPT-6 Sol on OSWorld 2.0 scored 60.5% against Claude Opus 5's 60.3% at roughly 80% less cost.

Anthropic, in turn, pointed to agentic coding benchmarks. On Terminal-Bench 4.0, which scores whether an agent can finish complex professional tasks inside a command-line interface as a percentage of tasks completed, Opus 5.5 hit 66.4%, ahead of Fable 5.1's 55.8% and GPT-6 Astra's 57.9%. On FrontierCode, which checks whether an agent's code changes would actually get 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 grades real-world professional work across 44 occupations using Elo, Opus 5.5 scored 1846, Fable 5.1 scored 1735 and Opus 5 scored 1708. Anthropic also acknowledged that GPT-6 Astra edged Opus 5.5 on AutomationBench, 41.4% to 40.0%, and noted gaps on Terminal-Bench-Science 0.1 as well.

GPT-6 Sol and Luna are available now in ChatGPT Work and Codex for Plus, Pro and Business tiers, alongside the existing GPT-6 Astra tier.