OpenAI cuts GPT-5.6 prices

OpenAI cut prices for two GPT-5.6 models on July 30 and added a faster, higher-priced processing option for its flagship Sol model, extending a pricing split between low-cost volume use and premium speed. The company said GPT-5.6 Luna now costs 80% less and GPT-5.6 Terra 20% less, while GPT-5.6 Sol Fast gives developers up to 2.5 times the speed of standard processing at double the standard price. The changes apply across OpenAI’s API and flow through to how usage is counted in Codex and some paid ChatGPT work products, according to OpenAI. The company said the lower prices reflect “improvements in how our models are built, served, and put to work,” rather than a change in model capability. ### How much did OpenAI actually cut the prices? (openai.com) OpenAI’s API pricing page lists GPT-5.6 Luna at $0.10 per million short-context input tokens and $0.60 per million short-context output tokens. The same page lists GPT-5.6 Terra at $1.00 for short-context input and $6.00 for short-context output, while GPT-5.6 Sol remains at $2.50 for short-context input and $15.00 for short-context output. (openai.com) OpenAI’s July 30 product post said Luna’s new rate is 80% below its prior level and Terra’s is 20% lower. The company described Luna as its “fastest and most affordable model” and Terra as its balanced model for everyday work. ### What is the new Sol fast option? OpenAI’s community announcement said GPT-5.6 Sol Fast delivers up to 2.5 times the speed of standard processing without changing the model’s intelligence. (developers.openai.com) The company said the faster option is aimed at time-sensitive API uses and is priced at twice the standard Sol rate. (openai.com) OpenAI’s broader GPT-5.6 launch materials describe Sol as the flagship model in the family, with Terra positioned for routine work and Luna for cost-efficient scale. The company has also been rolling GPT-5.6 Sol into eligible ChatGPT plans, though availability depends on account and plan type. ### Why does OpenAI say it can charge less now? OpenAI said the lower prices came from internal efficiency gains in inference and model serving. (community.openai.com) In its July 30 post, the company said the cuts reflect years of work on model construction, serving and deployment efficiency, and that it is “passing those gains on to customers.” (openai.com) The pricing move also coincides with product changes in developer tooling. OpenAI’s Help Center says GPT-5.4 and GPT-5.4 mini will retire from Codex for signed-in ChatGPT users on August 31, 2026, with GPT-5.6 Terra and GPT-5.6 Luna named as replacements. ### Did cheaper inference solve the security problem? OpenAI’s updated GPT-5.6 safety disclosures did not show a simple reduction in all attack risks. (openai.com) TechRepublic, citing the Aug. 3 update to OpenAI’s system card, reported that GPT-5.6 Sol failed on about 0.05% of direct prompt-injection attacks generated by OpenAI’s red-teaming system, but indirect attacks routed through tools and external content succeeded more often. (help.openai.com) TechRepublic said average indirect attack success rates reached 3.77% for Sol, 3.32% for Terra and 2.94% for Luna in OpenAI’s tests. The publication said those attacks can arrive through emails, webpages, uploaded files, code repositories and tool responses, and that the figures measure test success rates rather than the probability of a production breach. (techrepublic.com) ### What changes next for customers using these models? August 31, 2026 is the next named cutoff in OpenAI’s documentation. On that date, OpenAI says GPT-5.4 and GPT-5.4 mini will retire in Codex for users signed in with ChatGPT, while API customers using their own keys are not affected by that specific retirement notice. (help.openai.com) (techrepublic.com)