GPT-5.6 Luna

Ofox's discounted pricing page (updated 2026-08-21) lists GPT-5.6 Luna at $0.20 input / $1.20 output / $0.02 cached input per million tokens, placing it firmly in the "below vendor list" discount group versus OpenAI's $1/$6 list. Alongside Terra at $2.00/$12.00/$0.20 and Sol at $2.50/$15.00/$0.25, Luna remains the entr The post contextualizes the Luna rate against other discounted models on Ofox, including GLM-5.2/5.3 and the Doubao Seed 2.1 family, and clarifies that text models bill per million tokens whereas Seedance models bill per second of output. For developers choosing Luna, the practical implication is an 80% cut on the vend

Ofox's official tier guide for the GPT-5.6 family documents Luna as the cheap workhorse tier, positioned at $1 input / $6 output per million tokens on vendor list price — exactly half of Terra and one-fifth of Sol. Luna shares the family's 1M context window and 128K output cap with Sol and Terra, and is reachable via t The same post provides the operational decision frame for tiering: Luna is the right default when every request is roughly the same difficulty and cost dominates over capability, while a router that mixes Sol/Terra/Luna per request is justified when the workload spans capability levels. Routing per request with explici

OpenAI's official preview post dated June 26, 2026 introduces the GPT-5.6 series comprising three tiers: Sol as the flagship, Terra as a balanced everyday model, and Luna as the fast, affordable tier. The post states that Luna "brings strong capability at our lowest cost," positioning it as the cost-optimized option fo Evaluations shared at preview highlight improved agentic capabilities in coding, biology, and cybersecurity, with additional safety and preparedness results available in the system card. The post frames GPT-5.6 as a family with capability tiers rather than a single model, allowing routing by task depth. However, the ex

On August 19, 2026, Replit announced a new Free Mode powered by OpenAI's GPT-5.6 Luna model. The announcement was made by Replit on X and amplified by co-founder and CEO Amjad Masad, positioning the integration as part of a broader shift where agents make software cheaper to produce while making coding more expensive. Replit's Free Mode lowers the entry barrier to its agent-led product, leveraging Luna to enable natural-language-to-application workflows including code, design, and deployment. The move comes as Replit scales an AI-native platform reporting more than 50 million users and adoption across more than 85% of Fortune 500 co

Layer3 Labs’ guide identifies GPT-5.6 Luna as the fastest and cheapest of OpenAI’s three GPT-5.6 tiers, positioning it for summarization, drafting, classification, and routine automation. It explicitly notes that broader state-of-the-art claims for the GPT-5.6 family are OpenAI claims rather than independently verified The guide says OpenAI reports that Luna outperformed Claude Fable 5 on Agents’ Last Exam at an estimated cost per task nearly 99% lower. It also explains that this benchmark covers long, multi-step agentic tasks, while emphasizing that the reported price-performance result remains an OpenAI-sourced claim rather than an

Yahoo Finance and Quartz report OpenAI reduced GPT-5.6 Luna pricing by 80% and GPT-5.6 Terra by 20% on July 30, 2026. Luna dropped from $1 to 20 cents per million input tokens and from $6 to $1.20 per million output tokens. Sol, the most powerful model, retained its original pricing. OpenAI attributed the reductions to efficiency gains made during internal development of GPT-5.6, noting the model's ability to rewrite and optimize production code improved token generation, reducing end-to-end serving cost by 20% and increasing token-generation efficiency by over 15%. The cuts reflect enterprise relu

OpenAI's Deployment Safety Hub published an August 2026 update for the GPT-5.6 family that explicitly distinguishes two versions of GPT-5.6 Luna: a new "August" Luna replacing GPT-5.5 Instant as the default for ChatGPT Free and Go users, and the previously released "July" Luna still served via Codex and ChatGPT Work. T The August update also broadens ChatGPT access: Plus and Pro users receive an updated GPT-5.6 Sol with a reasoning-effort slider, while Free and Go users get the new Luna as default for everyday chats. The page is the first to include dedicated U18 (under-18) evaluations designed to measure model behavior against teen-

CNBC reports OpenAI slashed pricing for two of its GPT-5.6 models on July 30, 2026, roughly three weeks after their public release. GPT-5.6 Luna's cost was cut by 80%, dropping to 20 cents per million input tokens and $1.20 per million output tokens. GPT-5.6 Terra received a 20% reduction to $2 per million input and $1 OpenAI attributed the cuts to efficiency gains including the model's ability to rewrite and optimize production code and improve token generation, stating its strategy focuses on advancing capability and efficiency so each generation can accomplish more work at lower cost. The move reflects pressure from cost-sensitive

On July 30, 2026, OpenAI detailed price reductions across the GPT-5.6 family, with GPT-5.6 Luna priced 80% lower and Terra 20% lower, while GPT-5.6 Sol gained a new Fast mode in the API that delivers up to 2.5x faster speeds than Standard processing at twice the price. An Artificial Analysis Intelligence Index v4.1 cha The same OpenAI price-performance post notes that the lower Luna and Terra prices are also reflected in how usage counts against paid subscriptions when using Codex and ChatGPT Work, and that Fast mode requests tagged priority will automatically route to the new fast tier. The quoted executive commentary frames GPT-5.6

Neowin reports OpenAI announced GPT-5.6 Sol, Terra, and Luna would be publicly available for everyone starting July 9, 2026, following U.S. Department of Commerce approval after additional testing. GPT-5.6 Luna was positioned as the smallest and most cost-efficient model in the series, priced at $1 per million input to The article also notes OpenAI introduced more predictable prompt caching for the GPT-5.6 series, including support for explicit cache breakpoints and a 30-minute minimum cache life, helping developers reduce token usage. GPT-5.6 Sol was framed as the flagship for complex reasoning and agentic workloads, while Terra was

AWS announced on July 13, 2026, that OpenAI's GPT-5.6 Sol, Terra, and Luna models are now generally available on Amazon Bedrock. The announcement positions GPT-5.6 Luna as part of OpenAI's latest model series becoming accessible to AWS customers through the Bedrock platform. Making GPT-5.6 Luna available on AWS Bedrock provides developers and enterprises with an additional deployment channel for the model, expanding hosting options beyond OpenAI's own API. The page confirms general availability status but the supplied excerpt is dominated by cookie consent text rather than detailed technic

On 26 June 2026, OpenAI announced three new models—GPT-5.6 Sol, Terra, and Luna—entering a limited restricted preview accessible only to approved partners via the API and Codex, with ChatGPT excluded and no public waitlist. Thirteen days later, on 9 July 2026, OpenAI declared the GPT-5.6 family generally available acro The Medium piece positions Luna as one of three capability tiers released simultaneously with Sol (flagship) and Terra (lower-cost), forming the GPT-5.6 family. The article separates the preview announcement from the GA rollout, noting that pricing, safety posture, and family structure were established during the previ