Anthropic releases Claude Opus 5.5 and OpenAI counters with two cheaper GPT-6 models
SiliconANGLE Duncan Riley ● Covered by 3 sources
Anthropic cut Claude Opus 5.5 prices and OpenAI fired back with two cheaper GPT-6 models. The race is now about cost as much as capability.
Based on reporting by SiliconANGLE, Duncan Riley — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic and OpenAI just showed how little the AI model wars have slowed, even with all the talk about runaway systems. Anthropic released Claude Opus 5.5 and trimmed pricing, then minutes later OpenAI answered with two new GPT-6 models, Sol and Luna, priced well below their predecessors.
Anthropic says Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, which it says is 40% cheaper on a typical workload than Opus 5. Cache reads fell hardest, down 60% to 20 cents. There’s also a faster serving mode, but it costs more: $8 per million input tokens and $40 per million output.
The company is making a simple pitch: this model is stronger, faster, and less chatty. Anthropic says Opus 5.5 matches Fable 5.1 on most work, while output generation is more than 30% faster than Opus 5. It also cites a 680,000-line code migration finished in under a day, something it says would have taken an engineering team weeks.
The benchmark numbers are broad, and they tilt in Anthropic’s favor. Opus 5.5 scored 66.4% on Terminal-Bench 4.0, 40% on AutomationBench, 58.7% on Terminal-Bench-Science 0.1, and 1846 on GDPval-AA v2.1 versus 1708 for Opus 5. On Humanity’s Last Exam, run with tools, it reached 67.7%. Anthropic also says safety improved, with containment-bypass attempts down 85% from Opus 5 and prompt-injection resistance matching Fable 5.1 on Gray Swan.
OpenAI’s answer was cheaper, not bigger. Sol and Luna are built on the same training methods as GPT-6 Astra, which began rolling out Sept. 3, but tuned for cost. Sol is aimed at coding and agent work; Luna is for high-volume routine jobs like summarization and extraction. OpenAI says Sol hits 33.2% on AutomationBench and 68.8% on DeepSWE v1.1, while Luna reaches 66.6% on DeepSWE v1.1. Neither launch gives the company a direct head-to-head with Anthropic’s new model, which is probably the point.
My take — AI-written commentary, not fact-checked reporting
This is the AI industry in one neat little cartoon: slower safety debate, faster pricing war. The smart money isn’t on whoever yells “frontier” loudest; it’s on whoever makes agents cheap enough that companies stop caring about the brand name on the box.
Read more about this at: SiliconANGLE
