Featured image of post GPT-6 Sol/Luna Launch with 50% Price Cut, Claude Opus 5.5 Cuts Costs by 40%, Sparking API Price War

GPT-6 Sol/Luna Launch with 50% Price Cut, Claude Opus 5.5 Cuts Costs by 40%, Sparking API Price War

OpenAI and Anthropic simultaneously slash API prices while upgrading model capabilities, signaling a new phase in AI inference cost reduction.

Giants Clash: GPT-6 Sol/Luna and Claude Opus 5.5 Launch with 50% and 40% Price Cuts

OpenAI and Anthropic released new flagship AI models on September 22, with drastic API price reductions paired with performance upgrades:

  • OpenAI launched GPT-6 Sol and GPT-6 Luna, cutting API prices by 50% versus GPT-5.6 promotional rates; available to Plus and higher-tier users
  • Anthropic introduced Claude Opus 5.5, its first 5.5-family model, reducing typical task costs by 40% while boosting output speed over 30%
  • Luna is accessible via desktop app for Free/Go users; API model names are gpt-6-sol and gpt-6-luna

Models are live immediately, with platform users able to access them across ChatGPT Work and Codex (OpenAI) or Claude API (Anthropic).

Performance and Pricing Breakdown

GPT-6 Sol and Luna target distinct cost-performance tiers:

ModelInput/Output Price (per 1M tokens)Factuality error rateDeepSWE 1.1 peak score
GPT-6 Sol$2 / $10~50% of predecessor68.8%
GPT-6 Luna$0.1 / $0.5-66.6%
Claude Opus 5.5$4 / $20 / cache $0.2--

A counterintuitive detail: Anthropic previously called for slowing frontier model progress, yet it launched Opus 5.5 promptly. The company disclosed external evaluations by Frontier Design and METR, with security aligned to Fable 5.1 standards—showing a balancing act between commercial traction and caution.

Edge AI and Enterprise Integration Advance

Edge AI capabilities are also maturing: Qualcomm’s 6th-gen Snapdragon 8 Super Ultra CPU reaches 5GHz, and StepEdge-Omni 30B-MoE deployed on device drops memory requirements by over 50%, achieving 330+ tokens/s prefill and 28+ tokens/s decoding—enabling local agent tasks like email sorting and itinerary planning without cloud calls.荣耀’s Magic9 series and Robot Phone, launching September 28, feature Qwen Intelligence with 91.8% task accuracy on-device.

OPPO’s Find X10 series launched simultaneously: Pro Max offers triple 200MP cameras at 6799 yuan (12GB+256GB), Find X10 pairs 200MP main and periscope with 8000mAh battery at 5499 yuan, Find X10 E三摄 50MP at 4999 yuan—setting a new benchmark for high-end smartphone pricing.

Adoption Recommendations

  • Enterprise developers: GPT-6 Luna ($0.1/$0.5) delivers strong unit economics for high-frequency lightweight tasks; Opus 5.5 suits scenarios needing retention of professional capabilities amid cost pressure
  • Individual users: Free/Go users can try Luna via desktop; local deployment merits attention for privacy-sensitive workloads—6th-gen Snapdragon 8 devices already run 30B MoE models on-board
  • Hardware buyers: OPPO’s aggressive pricing at 6799 yuan with triple 200MP cameras makes Pro Max compelling if isolated tango 9600 Pro performance isn’t critical

In Conclusion

The AI pricing war signals an industry shift from “performance arms race” to “commercial sustainability,” with model capabilities now entering the practical adoption threshold—true competitive advantage will reside in maintaining service quality under cost constraints.

原文配图1
原文配图1|News screenshot

原文配图2
原文配图2|News screenshot

原文配图3
原文配图3|News screenshot

原文配图4
原文配图4|News screenshot