The current GPT model map
On November 18, 2026, the live catalog (GET /v1/models, which needs no key) returned eight GPT text IDs: gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra, gpt-6-sol, gpt-6-luna, gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna. This article adds prices from the live Claudexia price list of the same day and makes no benchmark claims. Prices and the model list change, so check the models page and the docs before you lock a configuration.
Per OpenAI's changelog, GPT-6.2 Astra and GPT-6.2 Sol were released on November 17, 2026, and OpenAI's model catalog calls GPT-6.2 Astra its flagship model for complex reasoning and coding. On November 18, 2026, the Claudexia price list has no GPT-6.2 models; the current list is on the models page. The comparison below covers only the GPT models that the price list does have.
Comparison table
Prices are in US dollars per million tokens, Premium channel, from the live Claudexia price list on November 18, 2026. The purpose column paraphrases OpenAI's own model pages and changelog.
| Model ID | Purpose per OpenAI | Input / output | First workload to test | What to measure |
|---|---|---|---|---|
gpt-6-astra | Reasoning model for complex coding and professional work | $0.78 / $3.90 | Hard prompts where a quality-first route is acceptable | Correctness on difficult cases and timeout rate |
gpt-6.1-sol | Near-Astra performance for complex work at a lower cost | $0.12 / $0.58 | General production tasks with mixed reasoning and writing | Task success, latency, and output length |
gpt-6.1-terra | Fast, efficient model for visual reasoning, computer use and clear writing | $0.24 / $1.20 | Text drafting and reading images or screenshots | Image-reading accuracy and writing style |
gpt-6-sol | Released on September 22, 2026 with Luna; not on OpenAI's recommended list | $0.38 / $1.86 | General coding, analysis, and structured drafts | Accuracy, schema compliance, and retries |
gpt-6-luna | Most efficient model for focused, high-volume tasks | $0.08 / $0.30 | Short classification and high-volume transformations | Throughput, parse failures, and cost in your account |
gpt-5.6-sol | Flagship of the GPT-5.6 family | $0.24 / $1.16 | Existing GPT 5.6 workloads that need a broad general route | Regression rate against your current fixtures |
gpt-5.6-terra | Balances intelligence and cost | $0.34 / $1.62 | Middle route between high quality and high throughput | Quality gain per added token and second |
gpt-5.6-luna | Cost-sensitive, high-volume workloads | $0.04 / $0.20 | Small extraction, labeling, and routing calls | Whether the task still passes without a retry |
All eight models have a context of up to 1,000,000 tokens in the Claudexia price list. Maximum output is 128,000 tokens, except gpt-6.1-terra and gpt-5.6-terra, which allow 32,000.
Two things stand out. On Claudexia gpt-6-sol costs about 3.2 times what gpt-6.1-sol does, while OpenAI charges the same for both in its own API: at the Standard tier $2 / $10 for a request of up to 272K input tokens and $4 / $15 above that threshold. And gpt-5.6-terra costs more than gpt-5.6-sol, while in OpenAI's API Terra is cheaper: up to 272K input tokens $2 / $12 against $4 / $20 (the Sol price is promotional until November 30, 2026), and above the threshold $4 / $18 against $8 / $30. The Claudexia price list gives no reason, and we do not know one.
The table is a test plan. It is not a ranking. A model that wins on a public coding example can lose on your support policy or JSON schema.
How to choose between GPT 6 and GPT 5.6
Start by writing down what failure costs. A typo in a draft title is cheap. A wrong migration plan, account action, or code review finding is not. Put the strictest checks on the expensive workflows, then test the least capable route against the same fixtures.
Keep old model IDs in a separate compatibility route while you migrate. This gives you a clear rollback target. Do not change the prompt, parser, and model in the same deploy. If the result changes, you will not know which variable caused it.
Build a small model harness
Use a fixture file with real anonymized requests. Each case should include the expected shape, a pass rule, and a note about what the model must refuse. Run the harness for every candidate with identical generation settings.
import OpenAI from "openai";
const openai = new OpenAI({
apiKey: process.env.CLAUDEXIA_API_KEY,
baseURL: "https://api.claudexia.tech/v1",
});
const candidates = [
"gpt-6.1-sol",
"gpt-6.1-terra",
"gpt-6-astra",
"gpt-6-sol",
"gpt-6-luna",
"gpt-5.6-sol",
"gpt-5.6-terra",
"gpt-5.6-luna",
];
for (const model of candidates) {
const result = await openai.chat.completions.create({
model,
messages: [
{ role: "system", content: "Return a JSON object." },
{ role: "user", content: fixture.prompt },
],
response_format: { type: "json_object" },
});
await saveResult({ model, result, fixtureId: fixture.id });
}
fixture and saveResult are your own code. Whether every model accepts response_format through the gateway, we did not check: if a call is rejected, drop the field and validate the JSON in your parser. Keep credentials in environment variables and never put a live key in the fixture repository.
Routing patterns that hold up
For a mixed application, send quick labels and simple transforms to gpt-6-luna or gpt-5.6-luna. Send general requests to gpt-6.1-sol, or to gpt-6-sol if your existing tests are pinned to it. Reserve gpt-6-astra for a queue with a clear reason to spend more time on a difficult case.
For code, route by risk. A testable one-file change can use a Sol or Luna model. A cross-service change should go through a quality-first route, then a human checks the diff and test output. The model does not replace the build.
A migration checklist
- Pin the current model ID in configuration.
- Save a fixture set from real, anonymized requests.
- Run the old and new models with the same prompt and parser.
- Compare pass rate, parse failures, latency, and token usage.
- Move a small traffic slice before changing the default.
The most useful dashboard separates model errors from application errors. A timeout, a schema failure, and an incorrect answer need different fixes.
Where to read next
The smaller GPT-5.6 models have their own articles: GPT-5.6 Luna and GPT-5.6 Sol and Terra.
FAQ
Which GPT model should I try first?
Start with gpt-6.1-sol for a mixed general workload. Test gpt-6-astra on the cases where correctness matters most, then test Luna on the simple queue.
Is GPT 6.1 Sol always better than GPT 5.6 Sol?
The model ID does not prove that. Run the same fixtures and keep the result with the better task score for your use case. On price, gpt-6.1-sol is cheaper on Claudexia on November 18, 2026: $0.12 / $0.58 against $0.24 / $1.16.
Can one API key call all GPT IDs?
A Premium-channel key can: on November 18, 2026, all eight IDs sit in Premium, and a Cheap-only key does not open them. Your account permissions and the live catalog remain the source of truth before a deploy. A gateway key is not an OpenAI key, and the difference, including OpenAI's intermediary rules effective 20 November 2026 (Usage Policies, section "Providing access to our models"), is explained in OpenAI API key: how to get one and OpenAI-compatible options. We do not give legal assessments and do not promise that a method is lawful or safe for your account. You read and accept the vendor's terms.
Should I use Luna for code review?
Use it only after a review fixture set shows acceptable findings and misses. For risky merges, pair the model with tests and a human approval step.
Where can I check current availability?
See the models documentation and the API response for the current catalog. Do not hard-code a model that is absent from that response.