Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
OpenAI’s major new-model announcement is a three-model GPT-5.6 family, not a single undifferentiated release. GPT-5.6 Sol targets the hardest reasoning, coding and agentic work; GPT-5.6 Terra balances capability and cost; and GPT-5.6 Luna is designed for fast, high-volume workloads. OpenAI announced general availability on July 9, 2026, after previewing Sol on June 26.
The biggest post-launch change is economic: on July 30, OpenAI cut Terra and Luna API prices and replaced Priority Processing with paid Fast mode. Which model makes sense depends on task difficulty, request volume, latency, tools and whether you are using ChatGPT, Codex or the separately billed API.
The short version
GPT-5.6 is a portfolio strategy with three tiers:
- Sol: the flagship for complex professional work, advanced coding, research, cybersecurity and long-running agents.
- Terra: a middle option for general business work, moderate analysis and applications where Sol costs more than necessary.
- Luna: the lowest-cost, fastest tier for extraction, classification, routine generation, support automation and other high-volume jobs.
“GPT-5.6” identifies the generation; Sol, Terra and Luna identify capability tiers that can evolve independently. Access varies by subscription, product, account and rollout, so a model being generally available does not mean every ChatGPT user sees every selector.
OpenAI’s current product information is available in its launch announcement and developer model index.
GPT-5.6 models compared
| Model | Best starting point | Positioning | Standard API price (per 1M tokens) |
|---|---|---|---|
| Sol | Hard reasoning, advanced coding, science, security and multi-step agents | Highest capability | $5 input / $30 output |
| Terra | General professional applications and moderate analysis | Capability-cost midpoint | $2 input / $12 output |
| Luna | Routine, latency-sensitive and high-volume processing | Fastest and least expensive | $0.20 input / $1.20 output |
These recommendations reflect OpenAI’s stated positioning, not an independent ranking. Your own representative prompts are the reliable way to choose.
What GPT-5.6 Sol adds
OpenAI positions Sol for tasks where an incorrect answer or failed tool call is expensive: complex professional workflows, command-line coding, scientific and biological research, cybersecurity analysis, computer use, design-related work and long-running agentic tasks that coordinate several tools.
Sol includes higher reasoning-effort settings. OpenAI describes max and an ultra mode that can coordinate multiple agents or subagents. These are product and configuration options, not necessarily separate model weights. More reasoning can improve difficult work, but it can also increase latency and token consumption; it does not guarantee factual accuracy.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Why Terra and Luna may matter commercially
Terra is intended for teams that need substantial capability without paying Sol rates for every request. OpenAI describes it as competitive with GPT-5.5 at a lower price; that is an OpenAI characterization, not a universal independent finding. It is a plausible starting point for business drafting, moderate coding, document analysis and internal assistants.
Luna’s economics are aimed at scale. Classification, extraction, routine rewriting, customer-support triage and batch processing can involve millions of calls, making a small per-token difference significant. A low token price still does not eliminate costs for retries, orchestration, storage, monitoring, human review or tools.
Availability in ChatGPT, Codex and the API
ChatGPT
OpenAI says Plus, Pro, Business and Enterprise users can access Sol at medium and higher effort settings. Pro and Enterprise users can select Sol Pro for the highest-quality results on complex tasks. Free and Go users receive Terra in ChatGPT Work, while Plus, Pro, Business and Enterprise users can select among Sol, Terra and Luna in ChatGPT Work and Codex where their interface and quotas support them. max and ultra availability differs by product and plan.
Rollouts, quotas and labels can change. Check the model picker and the current ChatGPT plans rather than assuming that an API model is available in your account.
Codex
GPT-5.6 is available in Codex, with effort settings and ultra access depending on subscription. Codex usage may consume credits or plan quotas. A ChatGPT or Codex subscription is not an unlimited API allowance.
OpenAI API
Developers can use the three models through OpenAI’s API and Responses API, subject to account access and the current model IDs in the documentation. API usage is metered separately from ChatGPT subscriptions.
Technical limits and caching
OpenAI’s current Terra and Luna pages list approximately a 1.05-million-token context window, up to 128,000 output tokens, text and image input, text output, multilingual support, vision and tool capabilities such as functions, web search, file search and computer use where configured. Those pages list a February 16, 2026 knowledge cutoff.
A context limit is not a promise of uniform quality across a million tokens. Retrieval, summarization or chunking may be cheaper and more reliable than sending an entire corpus on every request.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallGPT-5.6 also introduces explicit cache breakpoints, a stated 30-minute minimum cache life, cache writes billed at 1.25 times the uncached input rate and cache reads discounted by 90 percent from the uncached input rate. Caching helps repeated prompt prefixes; it does not make arbitrary prompts cheap.
Pricing after the July 30 update
The launch prices were Sol $5/$30, Terra $2.50/$15 and Luna $1/$6 for input/output per million tokens. On July 30, OpenAI reduced Terra to $2/$12 and Luna to $0.20/$1.20. Sol’s standard price did not change. Verify volatile rates on the live API pricing page before committing to a budget.
| Model | Input | Output | Simplified 1M input + 1M output example |
|---|---|---|---|
| Sol | $5 | $30 | $35 |
| Terra | $2 | $12 | $14 |
| Luna | $0.20 | $1.20 | $1.40 |
The examples exclude cached-input discounts, long-context rules, tool calls, search, computer use, retries and other charges. Input and output are billed separately, and long answers can dominate a bill even when input is inexpensive.
Fast mode
OpenAI renamed Priority Processing to Fast mode on July 30. It is a lower-latency service tier, not a smarter model. OpenAI says Sol Fast mode can be up to 2.5 times faster and costs twice Sol’s standard processing price. Existing API requests using service_tier: "priority" remain backward-compatible. Parameter names and supported values can change, so follow the current documentation rather than treating an example configuration as permanent.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →What OpenAI claims about performance
OpenAI says Sol reaches a new high on its Agents’ Last Exam evaluation and improves on selected coding, knowledge-work, cybersecurity and science evaluations. Its efficiency analysis also reports comparable or better results with fewer tokens and lower estimated cost in some tests, and strong placement on the Artificial Analysis Intelligence Index.
Those are primarily OpenAI-reported results and selected evaluations. They do not prove that Sol wins every competitor comparison or every real production task. Test representative prompts, tool calls, structured outputs, latency and failure recovery on your own workload.
Safety and governance
The June preview of Sol was restricted to selected trusted partners and organizations while OpenAI engaged with the U.S. government about cyber capabilities and future release processes. OpenAI reports human red-teaming, automated testing, model-level safeguards, monitoring and real-time checks, with access calibrated to risk.
Cybersecurity capability raises both defensive value and misuse risk. OpenAI’s preview system card also acknowledges that no evaluation covers every product configuration, multi-step attack or real-world workflow. Benchmark results are not a safety guarantee. Production teams should add permission boundaries, logging, prompt-injection defenses, review gates and limits on autonomous actions.
Which GPT-5.6 model should you choose?
- Choose Sol when the task requires sustained reasoning, advanced coding, research, complex tool coordination or high-cost error reduction.
- Choose Terra for balanced production workloads where Sol’s extra capability is not consistently needed.
- Choose Luna for repeatable classification, extraction, routine generation, support workflows and very high request volume.
- Choose Fast mode only when latency directly affects users or revenue enough to justify its premium.
Before migrating, test old and new prompts, structured-output schemas, tool definitions, refusal behavior, context retrieval and cost. Watch for hallucinated citations, wrong-file or wrong-function tool calls, prompt injection, agent loops, long-context degradation, unsafe cyber advice, rate limits and unexpected output-token growth.
Why this announcement matters
The important story is a model portfolio and pricing strategy. OpenAI is offering a premium flagship, a midrange alternative and a very low-cost high-volume tier, plus a paid speed option. That makes GPT-5.6 relevant not only to people asking which model is “smartest,” but also to businesses calculating whether an AI feature can operate within its margins.
Availability, quotas, model aliases and prices are volatile. Recheck OpenAI’s product, model and pricing pages before publication or deployment, and treat the model that passes your own evaluation—not the one with the strongest headline benchmark—as the right choice.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errors
