Benchmark result Provisional 60% confidence first seen
Two benchmark reports compare Sol against Opus 5.5 on browser-agent performance and on end-to-end user jobs. Sol scored higher than Opus 5.5 at the medium setting on the Browser Use browser-agent benchmark, while a separate user benchmark indicates Opus was preferred on most usable jobs but took longer and cost more than Sol.
The daily briefing
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.
TLDRocket
23 September 2026
AI’s business logic got cheaper—and more tightly controlled—today. OpenAI cut API pricing for GPT-6 Sol and GPT-6 Luna by half versus GPT-5.6 Sol, with Luna’s output-token price down 58%, and it added more default prompt caching plus a Prompt Caching Dashboard to help developers stop burning budget on repeated context. Anthropic followed with a discount for Claude Opus 5.5, lowering input from $5 to $4 per million tokens and output from $25 to $20, then warned that four common agent behaviors tied to Opus 5.5 could break in ways that surface as 400 errors—mostly around tool-choice, routing, and how conversation history is handled.
Read the full briefing →