Anthropic and OpenAI each used new model launches on Sept. 22 to make a similar practical pitch: developers can get more capability without paying the older rate for every token. The companies announced Claude Opus 5.5, GPT-6 Sol and GPT-6 Luna, but the releases target different spots in a model stack rather than offering a simple one-for-one replacement.
Anthropic lists Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. Its listed cache-read rate falls from $0.50 to $0.20 per million tokens. Anthropic says its own default-setting tests show typical workloads costing 40% less and output generating more than 30% faster. The company says the model is available in Claude, Claude Code and the Claude Platform, with safeguards and verification programs for higher-risk biology and cybersecurity work.
A cheaper pair for different work
OpenAI says GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens. It positions Sol for complex coding and agentic workflows. GPT-6 Luna is the lower-cost option, listed at $0.10 for input and $0.50 for output per million tokens, for focused and high-volume tasks. OpenAI says both models are available in its API and in specified Work and Codex plans, while broader Chat availability is staged.
The published rates are only the opening comparison. Caching, context length, tool calls, effort settings and error rates can change the effective price of a real workflow. Anthropic's efficiency figures and OpenAI's capability comparisons are their own evaluations, and the releases do not supply a shared independent test across the three models.
For teams deciding what to run, that leaves the ordinary but useful next step: compare the price schedule with the task. A model that is cheaper per token may not be cheaper for a long-running agent if it needs more steps, while a higher-priced model can make sense when it reduces retries or handles a harder job. The same-day releases make that tradeoff more visible, not automatic.
Tech