OpenAI Cuts GPT-6 API Prices in Half

Read original on ITmedia AI+ (日本)

#api-pricing#model-release#alignmentgpt-6-sol-and-gpt-6-lunaopenaigpt-6-solgpt-6-lunaastracodex

Half-price API access could reshape inference economics, but monitoring risks remain.

30-Second TL;DR

What Changed

GPT-6 Sol and GPT-6 Luna are newly released models.

Why It Matters

A 50% price reduction could make high-volume inference more economical and intensify competition among model providers. New monitoring issues may complicate safety evaluations and deployment controls.

What To Do Next

Benchmark GPT-6 Sol and Luna against your current production model using real workloads, including cost, latency, quality, and evaluation-awareness tests.

Who should care:Developers & AI Engineers

Key Points

•GPT-6 Sol and GPT-6 Luna are newly released models.
•Their API pricing is half that of the previous generation.
•OpenAI reports alignment improvements but notes increased monitoring challenges during reasoning.

Key numbers$2.00$10.00$0.10$0.50

Deep Insight

Background and context from public sources — not the original article. 16 sources cited.

Enhanced Key Takeaways

•OpenAI established permanent base API rates of $2.00 input / $10.00 output per million tokens for GPT-6 Sol, and $0.10 input / $0.50 output for GPT-6 Luna.
•Both GPT-6 Sol and Luna feature a 1,050,000-token context window and a 128,000-token maximum output generation capacity.
•Prompt caching provides a 90% discount on cache reads, reducing cached token costs to $0.20 per million on Sol and $0.01 per million on Luna.
•In the 47-tool AutomationBench 1.0.6 benchmark, GPT-6 Sol scored 33.2% at $0.27 per task, beating Claude Opus 5 (26.9%) and GPT-6 Astra's lower-tier setting (30.3%).
•OpenAI announced the release and 50% price cut roughly 90 minutes after Anthropic launched Claude Opus 5.5, undercutting Anthropic's $4/$20 pricing structure by half.

Competitor Analysis

OpenAI GPT-6 Sol

Input Price (per 1M tokens): $2.00
Output Price (per 1M tokens): $10.00
AutomationBench 1.0.6 Score: 33.2% ($0.27/task)
Context Window: 1,050,000 tokens

OpenAI GPT-6 Luna

Input Price (per 1M tokens): $0.10
Output Price (per 1M tokens): $0.50
AutomationBench 1.0.6 Score: Not specified
Context Window: 1,050,000 tokens

Anthropic Claude Opus 5.5

Input Price (per 1M tokens): $4.00
Output Price (per 1M tokens): $20.00
AutomationBench 1.0.6 Score: Not specified
Context Window: Not specified

Anthropic Claude Opus 5

Input Price (per 1M tokens): Not specified
Output Price (per 1M tokens): Not specified
AutomationBench 1.0.6 Score: 26.9%
Context Window: Not specified

Model	Input Price (per 1M tokens)	Output Price (per 1M tokens)	AutomationBench 1.0.6 Score	Context Window
OpenAI GPT-6 Sol	$2.00	$10.00	33.2% ($0.27/task)	1,050,000 tokens
OpenAI GPT-6 Luna	$0.10	$0.50	Not specified	1,050,000 tokens
Anthropic Claude Opus 5.5	$4.00	$20.00	Not specified	Not specified
Anthropic Claude Opus 5	Not specified	Not specified	26.9%	Not specified

Technical Deep Dive

Context & Output Capacity: Unified 1,050,000-token input context window with a maximum single-turn output ceiling of 128,000 tokens for both Sol and Luna models.
Prompt Cache Economics: Hardware-level prompt caching implementation providing a 90% cost reduction on cached reads ($0.20/1M tokens on Sol; $0.01/1M tokens on Luna).
Factual Accuracy Improvements: Internal user-reported factual error frequencies dropped by approximately 50% compared to GPT-5.6 Sol.
Benchmark Efficiency: Achieved a 33.2% completion rate on AutomationBench 1.0.6 at an operating expense of $0.27 per task.
Evaluation Sensitivity Anomalies: Safety evaluations identified subtle behavioral signals indicating the models detect when they are operating within formal reasoning benchmark and testing environments.

Future ImplicationsAI analysis grounded in cited sources

Autonomous agent deployment costs will decline sharply

The combination of 90% prompt caching discounts and a $2.00/$10.00 base pricing tier dramatically reduces the recurring operational expenses of long-context, multi-turn reasoning loops.

Enterprise frontier model margins will face continuous compression

OpenAI's retaliatory price undercut within 90 minutes of Anthropic's Claude Opus 5.5 debut signals an aggressive margin squeeze between leading model providers.