Sol, Terra, and Luna, our GPT‑5.6 family of models, are starting to roll out now…

According to data from independent evaluation platform Artificial Analysis, GPT-5.6 Sol (max configuration) scored 59 on the Intelligence Index, just 1 point lower than Anthropic's Claude Fable 5 (60). On the Coding Agent Index, Sol topped with 80 points in OpenAI's own Codex runtime environment, surpassing Fable 5's 77 in Claude Code. Additionally, on the AA-Briefcase office task benchmark, Sol achieved the highest "Demo Elo" score, but its overall ranking still lags behind Fable 5.

OpenAI officially claims that Sol leads in coding, knowledge work, cybersecurity, and science. Notably, differences in benchmark environments may affect results: Sol's high score in Codex relies on a proprietary environment, while Fable 5 performs more consistently in general scenarios.

With similar performance but vastly different costs, this result directly challenges Anthropic's previously established high-end positioning with Fable 5.

GPT-5.6 Sol costs $1.04 per task, compared to $2.75 for Claude Fable 5 (max configuration), making Sol only 38% of the latter. The family's lower-priced models are even more aggressive: Terra costs $0.55 per task, with performance comparable to GPT-5.5; Luna costs just $0.21, targeting high-throughput scenarios. Artificial Analysis notes that Sol "defines a new Pareto frontier" between intelligence and output tokens per task.

In terms of pricing innovation, GPT-5.6 introduces cache write fees for the first time. Input/output token prices: Sol at $5/$30 per million tokens, Terra at $2.50/$15, Luna at $1/$6. Cache reads receive a 90% discount, further reducing actual usage costs. Altman says that Sol consumes up to 54% fewer output tokens than comparable models in agentic coding tasks, implying higher cost-effectiveness per task.

This pricing strategy directly responds to the industry's "price deflation" trend. Chinese open-source models, Meta's Muse 1.1, and xAI's Grok 4.5 are all forcing top vendors to cut prices. Even though OpenAI's own API prices are still higher than some competitors, it is squeezing Anthropic's space through differentiated pricing.

The release of the GPT-5.6 family marks a shift in AI model competition from pure performance metrics to cost-effectiveness. For developers, lower inference costs mean more application scenarios become viable; but for the industry as a whole, the price war may erode profits and affect long-term investment. Whether Anthropic will follow with price cuts remains unknown—if it does, it may damage its profit margins; if it doesn't, it may lose market share.

However, the one-sidedness of benchmarks requires caution. The Intelligence Index only reflects specific task sets, and real-world application performance may differ. Additionally, Sol's high scores partly depend on OpenAI's own Codex environment, raising questions about cross-platform fairness. OpenAI's claimed token savings have not been independently verified and are a unilateral statement from the CEO.

In the long term, the commoditization of models is irreversible. The next phase of competition may shift focus to ecosystem stickiness, toolchain integration, and deep optimization in vertical domains, rather than mere parameter or price battles.