--- title: OpenAI launches GPT-6 Sol and Luna at half the price of... description: OpenAI released GPT-6 Sol and GPT-6 Luna at half the API price of GPT-5.6, launched shortly after Anthropic's cheaper Claude Opus 5.5. Sol targets complex... canonical: https://daily.dev/posts/openai-launches-gpt-6-sol-and-luna-at-half-the-price-of-gpt-5-6-with-improved-prompt-caching-fpbulrvtp twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: OpenAI launches GPT-6 Sol and Luna at half the price of GPT-5.6, with improved prompt caching | daily.dev og:description: OpenAI released GPT-6 Sol and GPT-6 Luna at half the API price of GPT-5.6, launched shortly after Anthropic's cheaper Claude Opus 5.5. Sol targets complex... og:url: https://daily.dev/posts/openai-launches-gpt-6-sol-and-luna-at-half-the-price-of-gpt-5-6-with-improved-prompt-caching-fpbulrvtp og:image: https://api.daily.dev/og/posts/FPbUlrvTp.png og:image:alt: OpenAI launches GPT-6 Sol and Luna at half the price of GPT-5.6, with improved prompt caching og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # OpenAI launches GPT-6 Sol and Luna at half the price of GPT-5.6, with improved prompt caching **[Collections](https://daily.dev/sources/collections)** · 5 min read · 8 upvotes · 1 comments ## Summary OpenAI released GPT-6 Sol and GPT-6 Luna at half the API price of GPT-5.6, launched shortly after Anthropic's cheaper Claude Opus 5.5. Sol targets complex tasks like coding at $2/$10 per million input/output tokens, while Luna targets high-volume simple workloads at $0.10/$0.50. Both ship in ChatGPT Work and Codex. Alongside the launch, OpenAI overhauled prompt caching: cached tokens now get a 90% discount within a 30-minute reuse window, plus new tools including a caching dashboard, cache-miss diagnostics, explicit cache breakpoints, mid-conversation reasoning effort changes without breaking cache, and cache prewarming. ## Content OpenAI released GPT-6 Sol and GPT-6 Luna about 90 minutes after Anthropic shipped a cheaper Claude Opus 5.5, and the two launches kicked off a fresh round of price cuts. ## The prices Sol and Luna cost roughly half what their GPT-5.6 equivalents did. Sol is aimed at complex work like coding and runs $2 per million input tokens and $10 per million output tokens. Luna is aimed at high-volume simple tasks: $0.10 in, $0.50 out, which makes it one of the cheapest OpenAI models ever. Both are already in ChatGPT Work and Codex. Anthropic's side of the trade: Claude Opus 5.5 is priced at $4/$20 per million, a 20% cut from Opus 5, with a 60% cut on cached input. Anthropic says it matches Claude Fable 5.1 on most work. Simon's pricing table lines the whole field up (Grok 4.7, GPT-5.6 Terra/Sol/Astra, Fable 5.1), and it's worth a look if you're comparing. OpenAI also published cost-per-task benchmarks against Claude Opus 5 and Fable 5, claiming comparable or better scores at much lower cost, plus improvements on factuality and alignment metrics. Treat vendor-picked benchmarks with the usual skepticism. Analysts (Forrester, Greyhound Research, Gartner) read the whole thing as sustained price compression, with models sliding toward commodity status, partly under pressure from cheaper Chinese open-weight models. Their advice: judge cost per business outcome, not cost per token. They also note that cheap APIs don't erase the case for private or hybrid deployments when data sovereignty and security are on the line. ## Caching is the second lever The headline price cut is easy to see. The caching changes are where I'd look for savings. Long-running agents keep resending the same instructions, tool definitions, and earlier context. Prompt caching lets the API skip recomputing those prefixes. OpenAI reworked it for GPT-6: - **Higher hit rates by default.** More input tokens land in the cached tier, which carries a discount of up to 90%. Shared prefixes reused within a 30-minute window qualify. - **A Prompt Caching Dashboard** showing hit rates, input composition, and cached versus uncached token volume. - **Miss diagnostics** that tell you why a request missed the cache, such as changed tools or settings. - **Explicit cache breakpoints**, so you decide which prefixes get cached. - **Reasoning effort changes mid-conversation** without breaking the cache, and tool availability changes too. OpenAI recommends append-only updates for tool and instruction changes. - **Prewarming**, which prepares shared context ahead of user requests to cut latency. OpenAI says these changes have already cut the share of prompt tokens needing fresh processing by more than 50% across billions of requests over several months. One caveat, which @rohanpaul_ai put well: a good hit rate is still partly an application-design problem. GPT-6's improved caching doesn't guarantee good economics on its own. If you scramble your prompt prefix every turn, you'll pay full price. Stable instructions and tools up front, volatile content at the end. ## Early field reports A few data points from around the ecosystem: - **Cost versus detail in 3D.** The team behind AI/ML API ran one-shot 3D scene prompts (starting with an animated fish school) through Opus 5.5 and GPT-6 Sol. Across four scenes, Opus 5.5 produced more detailed renders for $4.37 total; Sol came in at $0.34. - **A rough pelican.** Simon's pelican-riding-a-bicycle test on Opus 5.5 at max thinking failed twice. Both runs hit the 128,000 output token limit while still reasoning, costing $2.56 and 20 minutes each. Cheaper per token doesn't help if the model burns the whole budget thinking. - **Luna for extraction.** @jeresig reports that gpt-6-luna (flex, low thinking) is excellent for text extraction, at half the cost of the next cheapest option and just as accurate. - **A bug, now fixed.** OpenAI fixed an image-encoding bug that was degrading image understanding in Sol and Luna. Visual tasks in the API and Codex, computer use included, should be back to normal. - **ARC-AGI.** ARC Prize posted verified results for Sol: 4.6% on ARC-AGI-3 at $5.6K with the standard harness, and 23.0% at $8.7K with the provider adapter harness. - **Pricing speculation.** @scaling01 called a GPT-6.1-Sol as competitively priced, and reasoned it should cost about 1.5x more to serve than Sol if it's just Sol with one loop. ## Where you can use them Both Sol and Luna are moving fast into developer tools. GitHub Copilot added Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7 across various plans, with Sol and Luna now generally available. The Copilot app also gained local sandboxing (public preview) to restrict agent access to files, networks, and credentials, plus OpenTelemetry support. Slack and Teams get mid-conversation model switching, JetBrains adds assisted approvals for agent sessions, and VS Code 1.139 adds Dev Container agent support. Warp and Framer added the new models as well. ## Routing between them With two tiers this differently priced, routing is the obvious next move. One tutorial builds a TypeScript ModelRouter for Node.js 22.6+ that picks between a Sol tier and an Opus 5.5 tier by cost, complexity, and latency health. The pieces: pre-inference token-cost estimation with tiktoken, a 0-100 heuristic complexity scorer, a streaming-aware circuit breaker (CLOSED, HALF-OPEN, OPEN), automatic fallback between tiers, and structured JSON telemetry through Pino. It closes with production notes on persistent budget state, rate-limit coordination, and adding more tiers. ## What I take from this The price war is real, but per-token price is the number vendors want you to look at. Sol at $2/$10 is cheaper than what came before; a 90% discount on cached input applied to a well-structured agent is a much bigger swing. And as the Opus 5.5 pelican runs show, a runaway reasoning budget can wipe out any per-token advantage. If you run agents, open the caching dashboard first, then compare models on what a finished task costs. ## Questions this post answers ### What is the pricing for OpenAI's GPT-6 Sol and Luna models? GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, targeting complex tasks like coding. GPT-6 Luna costs $0.10 input and $0.50 output per million tokens, aimed at high-volume simpler workloads. Both are priced at half the rate of their GPT-5.6 equivalents and are already available in ChatGPT Work and Codex. _Developers comparing GPT-6 pricing tiers against other models can track releases like this on daily.dev._ ### How does the new OpenAI prompt caching discount work for GPT-6? Cached input tokens that share a reused prefix within a 30-minute window now get a 90% discount, up from previous rates. The update also adds a caching dashboard for hit rates, a diagnostics tool explaining cache misses, explicit cache breakpoints, mid-conversation reasoning effort changes that preserve cache, and cache prewarming to cut latency. _Teams optimizing LLM costs around shared system prompts follow caching changes like this via daily.dev._ ### Why did OpenAI launch GPT-6 Sol and Luna right after Anthropic's Claude Opus 5.5 price cut? The GPT-6 launch came roughly 90 minutes after Anthropic shipped a cheaper Claude Opus 5.5, reflecting an intensifying AI price war. Analysts attribute the pressure partly to cheaper Chinese open-weight models pushing incumbents to lower costs while OpenAI simultaneously published benchmarks claiming comparable or better performance than Claude Opus 5 and Fable 5. _Developers deciding between competing model providers watch pricing moves like this through daily.dev._ ## Community take How the wider developer community reacted, aggregated from 1 discussion and 11 comments across x (as of 2026-09-29). **TL;DR:** Replies focus mostly on whether the API price cuts actually benefit subscription users, with several arguing subscriptions get worse treatment despite cheaper API pricing, alongside some pushback comparing Sol against Opus 5.5. **Sentiment:** 10% positive · 30% mixed · 60% skeptical **The case for** - Some argue the pricing strategy is a smart, sustainable move to retain developers against competing models. **The pushback** - Several claim the cheaper API pricing doesn't carry over to subscription plans, which they say still get nerfed or throttled. - One commenter alleges subscription usage gets deprioritized or throttled when API/enterprise traffic is high. - Another dismisses the new model as just a rebrand of an existing one. **By community** - x (heated): Replies argue over whether subscription users actually benefit from the price cuts, plus a side debate on how Sol compares to Opus 5.5. **Hottest debate:** Whether the cheaper API pricing actually benefits subscription users or if subscriptions are quietly deprioritized/throttled. **Open questions** - Is there real evidence that subscription usage pools get reduced when API traffic is high? - Does the new pricing meaningfully change how Sol/Luna compare to Opus 5.5 in practice? **Highlights** > @WinterArc2125 @FriesIlover49 @scaling01 These API price cuts don't apply to the subscription. The usage in subscriptions are based on (time) * (model coefficient) * (fast mode off (1) | on (1.5)) Some days you can burn 1 billion tokens from weekly usage and others it's only 50 million. Weekends have more usage. > — [virtualcity69 on x · 1 points, 1 comments](https://x.com/virtualcity69/status/2104993258984083767) > @WinterArc2125 @FriesIlover49 @scaling01 Servers are deferred to API users first. If OpenAI ever has their servers hit hard by enterprise API users and you're in the middle of subscription work, they're gonna reduce your usage pool (draw it down faster) and if it gets real bad, kick you out of your session. > — [virtualcity69 on x · 1 points, 1 comments](https://x.com/virtualcity69/status/2104995934643855662) > @virtualcity69 @WinterArc2125 @FriesIlover49 @scaling01 this is simply untrue, never in my one year of using codex with the most expensive plan have i been kicked out of session > — [insanowskyy on x · 1 comments](https://x.com/insanowskyy/status/2105005682730607072) > @Rob3rtWozny @scaling01 Opus doesn’t wreck Astra, they are on par, and Sol 6.1 costs half as much as Opus. It’s ok to be a fanboy but calma down lol > — [techguyagent on x · 3 points](https://x.com/techguyagent/status/2104998213233278979) > @scaling01 Also the efficiency gains in the API aren't in subs AGAIN. > — [FriesIlover49 on x · 20 points, 1 comments](https://x.com/FriesIlover49/status/2104987931819602383) **Source threads** - [x](https://x.com/scaling01/status/2104986773990330516) · 0 points · 11 comments ## Community discussion Top comments from developers on daily.dev. **@pdfopsdev** · 0 upvotes > the part that actually matters here is decoupling reasoning effort from the cached prefix. usually effort is baked into the request hash so changing it mid convo forces a full cache miss, so if they actually pulled that off it's a real architecture change, not just a price cut. --- Tags: [#ai-inference](https://daily.dev/tags/ai-inference), [#claude](https://daily.dev/tags/claude), [#gpt](https://daily.dev/tags/gpt), [#llm](https://daily.dev/tags/llm), [#openai](https://daily.dev/tags/openai) [View this post on daily.dev](https://daily.dev/posts/openai-launches-gpt-6-sol-and-luna-at-half-the-price-of-gpt-5-6-with-improved-prompt-caching-fpbulrvtp) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"OpenAI launches GPT-6 Sol and Luna at half the price of GPT-5.6, with improved prompt caching","url":"https://daily.dev/posts/openai-launches-gpt-6-sol-and-luna-at-half-the-price-of-gpt-5-6-with-improved-prompt-caching-fpbulrvtp","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/openai-launches-gpt-6-sol-and-luna-at-half-the-price-of-gpt-5-6-with-improved-prompt-caching-fpbulrvtp"},"datePublished":"2026-09-22T20:15:58.208Z","dateModified":"2026-10-14T02:22:54.067Z","description":"OpenAI released GPT-6 Sol and GPT-6 Luna at half the API price of GPT-5.6, launched shortly after Anthropic's cheaper Claude Opus 5.5. Sol targets complex...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/cabe69b261c04edca5bbc3ce26b42aae?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/cabe69b261c04edca5bbc3ce26b42aae?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/openai-launches-gpt-6-sol-and-luna-at-half-the-price-of-gpt-5-6-with-improved-prompt-caching-fpbulrvtp","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":8},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"ai-inference,claude,gpt,llm,openai","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"OpenAI launches GPT-6 Sol and Luna at half the price of GPT-5.6, with improved prompt caching"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/openai-launches-gpt-6-sol-and-luna-at-half-the-price-of-gpt-5-6-with-improved-prompt-caching-fpbulrvtp","comment":[{"@type":"Comment","text":"the part that actually matters here is decoupling reasoning effort from the cached prefix. usually effort is baked into the request hash so changing it mid convo forces a full cache miss, so if they actually pulled that off it’s a real architecture change, not just a price cut.","datePublished":"2026-09-23T18:05:08.155Z","url":"https://daily.dev/posts/FPbUlrvTp#c-i1P0PY27n","author":{"@type":"Person","name":"PDFops","url":"https://daily.dev/pdfopsdev","image":"https://media.daily.dev/image/upload/s---8isRBKc--/f_auto/v1782922291/avatars/avatar_orjMeK8QKaaVZwGq7ScPz?_a=BAMAMicg0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/openai-launches-gpt-6-sol-and-luna-at-half-the-price-of-gpt-5-6-with-improved-prompt-caching-fpbulrvtp#faq","mainEntity":[{"@type":"Question","name":"What is the pricing for OpenAI's GPT-6 Sol and Luna models?","acceptedAnswer":{"@type":"Answer","text":"GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, targeting complex tasks like coding. GPT-6 Luna costs $0.10 input and $0.50 output per million tokens, aimed at high-volume simpler workloads. Both are priced at half the rate of their GPT-5.6 equivalents and are already available in ChatGPT Work and Codex. Developers comparing GPT-6 pricing tiers against other models can track releases like this on daily.dev."}},{"@type":"Question","name":"How does the new OpenAI prompt caching discount work for GPT-6?","acceptedAnswer":{"@type":"Answer","text":"Cached input tokens that share a reused prefix within a 30-minute window now get a 90% discount, up from previous rates. The update also adds a caching dashboard for hit rates, a diagnostics tool explaining cache misses, explicit cache breakpoints, mid-conversation reasoning effort changes that preserve cache, and cache prewarming to cut latency. Teams optimizing LLM costs around shared system prompts follow caching changes like this via daily.dev."}},{"@type":"Question","name":"Why did OpenAI launch GPT-6 Sol and Luna right after Anthropic's Claude Opus 5.5 price cut?","acceptedAnswer":{"@type":"Answer","text":"The GPT-6 launch came roughly 90 minutes after Anthropic shipped a cheaper Claude Opus 5.5, reflecting an intensifying AI price war. Analysts attribute the pressure partly to cheaper Chinese open-weight models pushing incumbents to lower costs while OpenAI simultaneously published benchmarks claiming comparable or better performance than Claude Opus 5 and Fable 5. Developers deciding between competing model providers watch pricing moves like this through daily.dev."}}]} ```