GPT-6 Sol and Luna Launch: 50% Cheaper Than GPT-5.6

Introduction

OpenAI introduced GPT-6 Sol and GPT-6 Luna on September 22, 2026, about three weeks after launching its flagship GPT-6 Astra. Where Astra is positioned as OpenAI's most capable and most expensive model, Sol and Luna extend the same training methods to faster, cheaper tiers, and OpenAI is cutting their API prices by half compared with GPT-5.6 Sol and Luna's promotional pricing. According to TechCrunch, the release arrived about 90 minutes after Anthropic's own Claude Opus 5.5 launch the same day, underscoring how closely the two companies are now timing competing releases.

Feature Overview

Pricing is the headline change. GPT-6 Sol drops from GPT-5.6 Sol's promotional $4 input / $20 output per million tokens to $2 / $10, a 50% cut. GPT-6 Luna falls from $0.20 / $1.20 to $0.10 / $0.50 per million tokens, also 50% cheaper. OpenAI says the reduction comes from improvements in caching and inference efficiency that let it serve the models at lower cost, and it is passing those savings on to users and customers. OpenAI did not announce any pricing change for GPT-6 Astra alongside this release, which it says remains its best model "across the board" for the most demanding work.

On OpenAI's own benchmark comparisons, GPT-6 Sol at xhigh effort scores 33.2% on AutomationBench (a 47-tool business-workflow test) at $0.27 per task, ahead of Claude Opus 5 at max effort (26.9%) for about 9% of Opus 5's cost per task. On Agents' Last Exam, GPT-6 Sol at max effort scores 56.4%, which OpenAI says is above Claude Opus 5's highest score in that evaluation at 60% lower cost per task. On DeepSWE v1.1, a software-engineering benchmark, GPT-6 Sol at max effort scores 68.8%, within 1.1 points of Claude Fable 5's best reported score (69.9%) at roughly 80% lower cost per task; GPT-6 Luna scores 66.6% at max effort, which OpenAI describes as comparable to Claude Opus 5 and Fable 5 at medium effort, while costing 93% less per task than Opus 5 and 96% less than Fable 5. On computer use, GPT-6 Sol at xhigh effort scores 60.5% on OSWorld 2.0's offline set, close to Claude Opus 5 at medium effort (60.3%), again at roughly 80% lower cost.

OpenAI also says GPT-6 Sol makes about half as many factual mistakes as GPT-5.6 Sol on its internal factuality evaluation, "approaching Astra-level reliability at much lower cost," while GPT-6 Luna at higher effort levels matches GPT-5.6 Sol's factuality at roughly one-hundredth the cost. Alongside the model launch, OpenAI shipped an improved prompt-caching system for the whole GPT-6 family: cache discounts of up to 90% on reused input tokens now apply to eligible shared prefixes reused within a 30-minute window, plus a new Prompt Caching Dashboard and diagnostics tool for developers to track and fix cache misses.

Usability Analysis

Sol and Luna sit at different points on the same cost curve. OpenAI positions Sol for demanding-but-routine professional and coding work where iteration speed matters more than squeezing out the last few points of accuracy, and Luna for lighter, high-volume tasks. Both are available today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu accounts, with a gradual rollout through the day; Free and Go users can reach Luna through the desktop app. Neither model is yet available in OpenAI's consumer Chat product, and both are callable in the API as gpt-6-sol and gpt-6-luna. OpenAI also carried Astra's more concise, less jargon-heavy communication style over to both models, which the company says is most noticeable in technical and coding conversations.

Pros and Cons

Pros:

• API prices cut 50% for both Sol and Luna versus their GPT-5.6 promotional rates, with Sol at $2/$10 per million tokens and Luna at $0.10/$0.50
• On OpenAI's own comparisons, both models beat or match pricier Claude models (Opus 5, Fable 5) on several benchmarks at a fraction of the cost per task
• Factuality gains are substantial: GPT-6 Sol reportedly makes about half as many mistakes as its predecessor, and Luna at higher effort matches GPT-5.6 Sol's factuality at roughly 1% of the cost
• Improved default prompt caching (up to 90% discount on reused tokens) plus new dashboard and diagnostics tools reduce cost for long agent sessions
• Immediate availability in ChatGPT Work, Codex, and the API for most paid tiers, with free access to Luna via the desktop app

Cons:

• Cost savings are measured against GPT-5.6's promotional pricing rather than a clearly stated standard list price, which makes the "50% cheaper" framing dependent on the baseline chosen
• Neither model matches GPT-6 Astra, which OpenAI still calls its best model "across the board" for the most demanding work
• Most competitive benchmark figures come from OpenAI's own testing of both its models and competitors' models, without independent reproduction
• Not yet available in OpenAI's consumer Chat product, limiting access for users outside ChatGPT Work, Codex, or the API

Outlook

By pricing Sol and Luna as the everyday-use tier beneath Astra, OpenAI is betting that most agentic and coding workloads don't need Astra's full capability or cost, and that the caching and inference efficiencies behind this price cut can keep extending to future releases. The improved prompt-caching system is explicitly built for the kind of long-running, multi-hour agent sessions OpenAI says are already driving exponential growth in its own internal token usage. Whether Sol and Luna's benchmark claims against Claude's Opus and Fable models hold up under independent testing, and whether Anthropic or Google respond with comparable price cuts at their own mid-tier levels, will likely shape the next round of this pricing competition.

Conclusion

GPT-6 Sol and Luna aren't a new intelligence tier so much as a cost restructuring of the one OpenAI already had: the same GPT-6-era training methods, delivered at half the API price of their GPT-5.6 predecessors' promotional rates. For developers and teams running high-volume coding or business-workflow agents where GPT-6 Astra's cost was hard to justify, this is the more practical model to build on. Users who need Astra's top-end reliability and computer-use performance still have to pay for it separately, since OpenAI has kept that model's pricing and positioning unchanged.

Editor's Verdict

GPT-6 Sol and Luna Launch: 50% Cheaper Than GPT-5.6 earns a solid recommendation within the GPT space.

The strongest case for paying attention: API prices cut 50% for both Sol and Luna versus their GPT-5.6 promotional rates, with Sol at $2/$10 per million tokens and Luna at $0.10/$0.50. That alone raises the bar for what readers should expect in this space. Reinforcing that, on OpenAI's own comparisons, both models beat or match pricier Claude models (Opus 5, Fable 5) on several benchmarks at a fraction of the cost per task — practical value rather than just headline appeal. The broader signal worth registering is straightforward: pricing, not a new capability tier, is the story here — Sol and Luna reuse Astra's training methods but exist mainly to move OpenAI's cost curve down, cutting API prices in half versus GPT-5.6's promotional rates rather than introducing new features. On the other side of the ledger, one constraint is real rather than a marketing footnote: cost savings are measured against GPT-5.6's promotional pricing rather than a clearly stated standard list price, which makes the "50% cheaper" framing dependent on the baseline chosen. It should factor into any serious decision. Layered on top of that, neither model matches GPT-6 Astra, which OpenAI still calls its best model "across the board" for the most demanding work — which narrows the set of teams for whom this is an obvious yes.

For ChatGPT power users, OpenAI API customers, and enterprise teams already running on the OpenAI stack, this is a serious evaluation candidate, not just a curiosity to bookmark. For everyone else, the safer posture is to monitor coverage and revisit once the use cases that matter to your team are demonstrated in the wild.