Claude Opus 5.5 Launches at 40% Lower Cost as OpenAI Fires Back

Anthropic released Claude Opus 5.5 on September 22, slashing API costs by 40% while delivering performance matching its premium Fable 5.1 model. Input pricing dropped to $4 per million tokens and output to $20, with cache reads falling 60% to $0.20. OpenAI responded within 90 minutes, launching GPT-6 Sol and Luna at half the price of their predecessors. The rapid-fire releases signal intensifying competition as both labs race to make frontier AI models affordable at scale.

Opus 5.5 cuts costs while matching top-tier performance

Claude Opus 5.5 delivers Fable 5.1-level performance on most tasks for 40% less than Opus 5, according to Anthropic’s announcement. Input costs dropped to $4 per million tokens and output to $20, with faster serving at $8 and $40. Cache reads plummeted 60% to $0.20 per million tokens. Output generation runs over 30% faster than Opus 5, which launched in late July.

On Terminal-Bench 4.0, an agentic coding test, Opus 5.5 scored 66.4% versus 55.8% for Fable 5.1. One early tester completed a 680,000-line code migration in under a day, work Anthropic said would take an engineering team weeks. On AutomationBench, the model achieved 40% task completion compared to 26.9% for Opus 5. In scientific research benchmarks, Terminal-Bench-Science 0.1 gave Opus 5.5 a score of 58.7% against 29% for Opus 5.

OpenAI launches cheaper Sol and Luna models 90 minutes later

OpenAI expanded its GPT-6 lineup on September 22 with Sol and Luna models priced at 50% of their GPT-5.6 predecessors. Sol now costs $2 per million input tokens and $10 per million output, while Luna sits at $0.10 and $0.50. OpenAI attributed the price cuts to efficiency gains in caching and inference. The company claims GPT-6 Sol makes about half as many mistakes as its predecessor on factuality evaluations and has lower error rates for coding tasks.

Safety testing and real-world performance drive adoption

Anthropic tested Opus 5.5 with external evaluators including Frontier Design and METR before release. The model scored highest on Anthropic’s automated behavioral audit across thousands of simulated scenarios. It resists prompt injection better than Opus 5 and is less likely to take hard-to-reverse actions or exceed its boundaries. Box Inc. VP Yashodha Bhavnani noted answers were 40% less verbose without losing accuracy. GitHub CPO Mario Rodriguez said the model solved more terminal tasks than Opus 5 in less than half the steps in VS Code.

Market implications for AI infrastructure costs

The price war cuts the cost of running frontier models at scale, making AI agents and automation economically viable for more use cases. Both companies are now prioritizing efficiency and affordability over raw capability gains. Opus 5.5 is now the default model in Claude Code, the Claude app, and Cowork for Pro, Max, and Team users. OpenAI rolled out Sol and Luna across ChatGPT Work, Codex, and the API, with Luna available to free users on desktop.

Final Thoughts

Anthropic and OpenAI are competing on cost and speed, not just raw power. Opus 5.5’s 40% price cut and OpenAI’s 50% discount signal the frontier AI market is shifting toward efficiency. For developers and enterprises, this means cheaper inference and faster iteration on agentic tasks.

FAQs

Disclaimer:

The content shared by Meyka AI PTY LTD is solely for research and informational purposes.  Meyka is not a financial advisory service, and the information provided should not be considered investment or trading advice.