# GPT-6 models compared: Astra, Sol, Terra and Luna

> OpenAI's GPT-6 family has five API models: GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol, GPT-6.1 Terra and GPT-6 Luna. GPT-6.1 Sol replaces GPT-6 Sol at the same price, and GPT-6.1 Terra comes close to it for half. Prices, scores and plan access as of October 29, 2026.

- Published: 2026-10-29
- Last reviewed: 2026-10-29
- Data as of: 2026-10-29
- Article slug: gpt-6-models-compared
- Author: [Glevd](https://x.com/glevd)
- Reading time: 6 minutes
- Topics: openai, gpt-6, comparison, pricing, api, chatgpt
- Data policy: Dated analysis; static benchmark claims are retained.
- Canonical URL: https://benchlm.ai/blog/posts/gpt-6-models-compared

OpenAI's GPT-6 family has five API models as of October 29, 2026, and two of the choices between them are easy.

GPT-6.1 Sol replaces GPT-6 Sol at the same price, and it scores higher on every test where we hold both.

GPT-6.1 Terra, released October 26, comes within a few points of GPT-6.1 Sol for half the price. It also guesses wrong less often than any other model in the family.

Per million input and output tokens, GPT-6 Astra costs $10/$50, GPT-6.1 Sol and GPT-6 Sol cost $2/$10, GPT-6.1 Terra costs $1/$5, and GPT-6 Luna costs $0.1/$0.5. All five list a 1,050,000-token context window. Scores below come from [Artificial Analysis](https://artificialanalysis.ai) (read October 28) and from OpenAI's own launch charts, labeled as such.

## GPT-6.1 Sol replaced GPT-6 Sol at the same price

OpenAI shipped GPT-6 Sol on September 22 and [GPT-6.1 Sol](https://openai.com/index/introducing-gpt-6-1-sol/) a week later, on September 29. Per-token prices didn't move, but cached input fell from $0.2 to $0.1 per million tokens.

| Model | Released | Price per 1M (input/output) | AA Intelligence Index | AA hallucination rate | DeepSWE v1.1 (OpenAI, max effort) |
| --- | --- | --- | --- | --- | --- |
| GPT-6 Astra | Sep 3 | $10/$50 | 56.9 | 49.8% | 74.1% |
| GPT-6.1 Sol | Sep 29 | $2/$10 | 54.5 | 51.9% | 71.9% |
| GPT-6 Sol | Sep 22 | $2/$10 | 47.6 | 60.1% | 68.8% |
| GPT-6.1 Terra | Oct 26 | $1/$5 | 51.0 | 34.5% | Not reported |
| GPT-6 Luna | Sep 22 | $0.1/$0.5 | 27.8 | 79.3% | 66.6% |

Artificial Analysis builds its Intelligence Index from its own evaluations, and higher is better. Its hallucination rate comes from a knowledge test: when a model doesn't have the right answer, it's the share of the time the model gives a wrong one instead of declining. Lower is better there.

Between the two Sols, the gap is wide on the independent index (54.5 against 47.6) and narrower on OpenAI's coding chart (71.9% against 68.8%). One detail in that chart: OpenAI reports GPT-6.1 Sol at 75.2% on high effort, above its 71.9% at max. Max effort costs more per task and isn't always the better setting.

Our pricing tables already lead the Sol tier with GPT-6.1 Sol, and we'd move any GPT-6 Sol code over for the same reasons. Run your own prompts first if you depend on exact output formats, because a new snapshot can change them.

## GPT-6.1 Terra makes the fewest wrong guesses for half the price

Terra sits between Sol and Luna on price. At $1/$5, it's half of GPT-6.1 Sol's rate, and Artificial Analysis scores it 51.0 against Sol's 54.5.

Its hallucination rate is the number to look at. At 34.5%, it's the lowest in the family, against 51.9% for GPT-6.1 Sol and 49.8% for Astra. For extraction, support answers or anything that has to say "I don't know" rather than invent a detail, Terra is the first GPT-6 model we'd test.

OpenAI's own agent charts put it level with the top tier. On Agents' Last Exam v1 at max effort, [OpenAI's Terra launch page](https://openai.com/index/introducing-gpt-6-1-terra/) reports GPT-6.1 Terra at 60.0%, GPT-6 Astra at 59.3% and GPT-6 Sol at 56.4%. It doesn't report GPT-6.1 Sol on that test. Those are OpenAI's runs on OpenAI's harness, so treat them as a reason to test, not as an independent result.

Terra is three days old as we write this. Our [GPT-6.1 Terra page](/models/gpt-6-1-terra) will fill in as more independent results arrive.

## Astra and Luna sit at the two ends for different reasons

GPT-6 Astra leads the family on most of the scores we hold, but not by much. Its Artificial Analysis index is 56.9 against GPT-6.1 Sol's 54.5, and OpenAI's DeepSWE chart has it at 74.1% against 71.9%.

At $10/$50, it costs five times GPT-6.1 Sol per token. A gap of 2.4 index points rarely pays for that on routine work. It can pay on the hardest tasks, where one failed run costs more than the tokens, so try Astra on the jobs where GPT-6.1 Sol fails, not on everything.

Astra is also the model OpenAI puts behind its own agents. OpenAI names GPT-6 Astra as the model running Dots, and the Pro 500 plan adds an Ultrafast mode for Astra and GPT-6.1 Sol.

At the other end, GPT-6 Luna costs $0.1/$0.5, a twentieth of GPT-6.1 Sol. OpenAI describes it as a model for focused, high-volume tasks.

OpenAI's coding chart flatters it: 66.6% on DeepSWE against GPT-6 Sol's 68.8%. Independent numbers are less kind. Artificial Analysis scores it 27.8, and its 79.3% hallucination rate is the highest in the family.

That makes Luna a fit for classification, routing, short rewrites and other work where the answer is checked or the input carries the facts. Don't use it as a source of facts on its own.

## ChatGPT plans don't carry every GPT-6 model the same way

API access and ChatGPT access don't line up one to one. [OpenAI's Codex pricing page](https://learn.chatgpt.com/docs/pricing), read October 16, lists GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna in Codex for Plus at $20 a month, and notes that GPT-6.1 Sol is available in ChatGPT Work and Codex, not Chat. Free and Go get GPT-6 Luna in Codex.

GPT-6.1 Terra reached ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users on October 26. The [ChatGPT plans page](/chatgpt/pricing-plans) keeps each plan's models and Codex estimates with OpenAI's sources.

For per-token rates across every OpenAI model, including the GPT-5.6 rows still on sale, see [OpenAI API pricing](/openai/api-pricing). Head-to-head pages such as [GPT-6.1 Sol vs GPT-6 Astra](/compare/gpt-6-1-sol-vs-gpt-6-astra) and [GPT-5.6 Sol vs GPT-6.1 Sol](/compare/gpt-5-6-sol-vs-gpt-6-1-sol) carry every benchmark we hold for each pair.

## Questions

### Is GPT-6.1 Sol better than GPT-6 Sol?

On every shared score we hold, yes, and it costs the same per token. Artificial Analysis puts GPT-6.1 Sol at 54.5 on its Intelligence Index against 47.6, and OpenAI reports 71.9% against 68.8% on DeepSWE at max effort. Cached input is also half the price, so there's little reason to stay on GPT-6 Sol.

### Which GPT-6 model is cheapest?

GPT-6 Luna, at a twentieth of GPT-6.1 Sol's per-token price. OpenAI built it for focused, high-volume work. On Artificial Analysis's knowledge test it gave a wrong answer instead of declining 79.3% of the time it lacked the right one, so check its output where a made-up fact would cost you.

### Which GPT-6 models come with ChatGPT Plus?

OpenAI's Codex pricing page, read October 16, lists GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna in Codex for Plus at $20 a month, with GPT-6.1 Sol in ChatGPT Work and Codex but not Chat. GPT-6.1 Terra reached Work and Codex for Plus on October 26.