GPT-5.6 Sol and Luna Update: What Changed in ChatGPT, What Did Not Change in the API, and How to Choose a Model

On August 6, OpenAI announced two ChatGPT updates around GPT-5.6.

First, GPT-5.6 Sol in ChatGPT was updated for Plus and Pro users, with a new reasoning-effort slider for everyday chat. Second, GPT-5.6 Luna is being rolled out as the default model for Free and Go users, with unlimited text conversations starting the following week, subject to abuse-prevention systems and unchanged limits on non-text features.

This update matters if you are:

• still using GPT-5.5 or older models
• choosing between Sol, Terra, and Luna
• running high-volume API workflows
• comparing GPT-5.6 with lower-cost alternatives such as DeepSeek
• deciding whether ChatGPT’s free Luna rollout changes your need for API access

The short answer:

ChatGPT users get a better Sol chat experience and broader Luna access.
API users should still evaluate GPT-5.6 by model ID, price, latency, and workflow fit.

ChatGPT users get a better Sol chat experience and broader Luna access.
API users should still evaluate GPT-5.6 by model ID, price, latency, and workflow fit.

The three GPT-5.6 models

OpenAI positions the GPT-5.6 family as three models with different roles: Sol, Terra, and Luna.

GPT-5.6 Sol

Sol is the flagship model.

The August 6 update improves the version of Sol used in everyday ChatGPT chat for Plus and Pro users. OpenAI says the new chat Sol is more concise, supports quick and deep reasoning through a slider, and reduces factual-error rate by about 68% versus GPT-5.5 Instant in OpenAI’s internal finance, medical, and legal evaluations.

Important boundary:

This August 6 Sol update applies to ChatGPT's everyday chat experience.
OpenAI says the Sol version powering Work and Codex is not changing as part of this release.

This August 6 Sol update applies to ChatGPT's everyday chat experience.
OpenAI says the Sol version powering Work and Codex is not changing as part of this release.

That means API users should not assume the August 6 ChatGPT chat behavior automatically changes gpt-5.6-sol API behavior. API pricing and capabilities should be checked against the current API documentation or your provider console.

OpenAI’s July 30 price-performance update also described Sol Fast mode, which offers higher speed at a higher price multiplier.

GPT-5.6 Terra

Terra is the general-purpose workhorse.

It is positioned for daily development, writing, content generation, and data-processing tasks where Sol may be more expensive than necessary but Luna may be too small for the job.

The July 30 OpenAI update says Terra’s pricing was reduced by 20%. In that same announcement, OpenAI included customer examples such as Notion using Terra for low-latency agent tasks and daily Q&A-style work.

For many teams, Terra is the safest first benchmark model:

Use Terra as the default test model.
Move up to Sol when quality requires it.
Move down to Luna when volume and cost dominate.

Use Terra as the default test model.
Move up to Sol when quality requires it.
Move down to Luna when volume and cost dominate.

GPT-5.6 Luna

Luna is the low-cost model in the family, but it is not just a simple small model.

OpenAI describes Luna as capable of tool use, long-context handling, and multi-step workflows. In the August 6 announcement, OpenAI says Luna reduced factual-error rate by about 62% versus GPT-5.5 Instant in internal evaluations.

The biggest user-facing change is ChatGPT availability:

• Luna rolls out as the default model for Free and Go users.
• Text conversations become unlimited the following week.
• Non-text features such as file uploads and image-related features keep their existing limits.
• Usage remains subject to abuse-prevention systems.

This is useful for casual ChatGPT users. It does not remove the need for API access if you are building an application, agent, workflow, or internal tool.

What is unlimited, and what is not

“Unlimited” is easy to misread.

The August 6 announcement is about unlimited text conversations in ChatGPT for Free and Go users after the Luna rollout. It does not mean unlimited API usage.

It also does not mean every ChatGPT feature becomes unlimited.

If you are building an app or agent, you still need API access.

Sol in ChatGPT: one model, adjustable effort

OpenAI’s August 6 post frames the Sol update as a fix for a familiar product issue: fast and deep modes often felt like different personalities.

The new ChatGPT Sol experience unifies the response style while letting users adjust reasoning depth.

In the official example, OpenAI compares an older GPT-5.5 Instant answer with the new GPT-5.6 Sol answer for a weather-related question. The newer answer gives the conclusion earlier and avoids unnecessary detail.

The product direction is clear:

Less generic verbosity.
More direct answers.
Adjustable reasoning effort.

Less generic verbosity.
More direct answers.
Adjustable reasoning effort.

For API builders, the equivalent lesson is not “always use Sol.” It is to choose effort and model tier based on task value.

Pricing context: official and apito.ai

OpenAI’s July 30 price-performance update reduced GPT-5.6 pricing, including an 80% reduction for Luna and a 20% reduction for Terra.

The source draft uses the following post-cut official pricing:

Through apito.ai, the source draft describes a 30% discount policy at publication time:

Treat provider pricing as time-sensitive.

Always verify the live apito.ai console before using these numbers for procurement, billing, or customer pricing.

Example monthly cost: support agent

Assume a support agent handles:

• 5,000 conversations per day
• 150,000 conversations per month
• 1,200 input tokens per conversation
• 600 output tokens per conversation

Using official GPT-5.6 Luna pricing from the source draft:

Using the apito.ai 30% discount assumption:

For high-volume agents, Luna is the model to test first. Move up to Terra or Sol only when task quality, reasoning, or tool reliability requires it.

GPT-5.6 Luna vs DeepSeek on cost

Luna’s 80% price cut is meaningful, but it does not automatically make Luna the cheapest model for every task.

DeepSeek’s official pricing page lists lower per-token prices for some models, especially when cache discounts apply. The source draft compares DeepSeek’s low-cost options with Luna and concludes that DeepSeek can still be cheaper for cost-sensitive, lower-complexity tasks.

That is a reasonable selection pattern:

DeepSeek’s pricing page also indicates that pricing may change, so do not hard-code a procurement decision from one snapshot.

Should you migrate from GPT-5.5?

If you are still using GPT-5.5 or older models, GPT-5.6 is worth testing.

OpenAI’s reported factual-error reductions for Sol and Luna are significant, especially for tasks involving:

• dates
• numbers
• citations
• finance
• medical content
• legal summaries
• customer-facing answers

But do not migrate blindly.

Use a small traffic split first.

Migration checklist

The common migration failure is not usually an SDK error.

It is a hidden prompt assumption. If your downstream logic expects a very specific format, run end-to-end tests before full rollout.

How to call GPT-5.6 through apito.ai

Prerequisites

• apito.ai account
• API key
• Python 3.8+ or Node.js 18+

API settings

Python example

from openai import OpenAI

client = OpenAI(
    api_key="your-api-key",
    base_url="https://gw.apito.ai/v1"
)

response = client.chat.completions.create(
    model="gpt-5.6-sol",
    messages=[
        {"role": "user", "content": "Explain speculative decoding in one sentence."}
    ]
)

print(response.choices[0].message.content)

from openai import OpenAI

client = OpenAI(
    api_key="your-api-key",
    base_url="https://gw.apito.ai/v1"
)

response = client.chat.completions.create(
    model="gpt-5.6-sol",
    messages=[
        {"role": "user", "content": "Explain speculative decoding in one sentence."}
    ]
)

print(response.choices[0].message.content)

Node.js example

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "your-api-key",
  baseURL: "https://gw.apito.ai/v1"
});

const response = await client.chat.completions.create({
  model: "gpt-5.6-luna",
  messages: [
    { role: "user", content: "Explain speculative decoding in one sentence." }
  ]
});

console.log(response.choices[0].message.content);

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "your-api-key",
  baseURL: "https://gw.apito.ai/v1"
});

const response = await client.chat.completions.create({
  model: "gpt-5.6-luna",
  messages: [
    { role: "user", content: "Explain speculative decoding in one sentence." }
  ]
});

console.log(response.choices[0].message.content);

Common errors

About Astra: treat it as unconfirmed

The source draft includes speculation about a possible next-generation OpenAI model called Astra.

Here is the careful version:

• OpenAI’s August 1 mathematics post says results were obtained by an internal version of Astra, described as the company’s next major model.
• Social-media posts have claimed that a later internal checkpoint, sometimes described as mewfour, may be close to release.
• Claims about parameter count, exact launch timing, GPT-6 naming, and comparisons to Anthropic Fable or Mythos are not officially confirmed.

Do not delay a GPT-5.6 migration solely because of Astra rumors.

If and when OpenAI publishes an official Astra release, reassess then.

FAQ

Which GPT-5.6 model should I choose?

Use Sol for complex reasoning and high-value coding, Terra for daily general work, and Luna for high-volume or cost-sensitive workflows. If you are unsure, benchmark Terra first.

If Luna is free in ChatGPT, do I still need the API?

Yes, if you are building an app, agent, automation, or internal workflow. ChatGPT’s free Luna text access is a consumer product feature. API usage is still separate and billed through API access.

What is different when using apito.ai?

apito.ai provides third-party API access with an OpenAI-compatible calling style, usage management, and provider-side pricing. It is not affiliated with OpenAI. Always verify live model availability and pricing in the apito.ai console.

Does the August 6 Sol update affect API Sol?

OpenAI says the August 6 update does not change the Sol version powering Work and Codex. Treat the update as a ChatGPT chat experience change unless your API provider documents otherwise.

Is Sol Fast mode worth using?

Use Fast mode when latency is worth the higher price, such as interactive agents or real-time user workflows. For batch processing, standard mode is usually more cost-efficient.

Should I wait for Astra?

No, not for production planning. Astra has been mentioned by OpenAI as an internal next major model in a mathematics post, but public product details are not final. Build against available models and keep your routing layer flexible.

Sources

• OpenAI: Improving GPT-5.6 Sol in ChatGPT
• OpenAI: Advancing the price-performance frontier with GPT-5.6
• OpenAI: GPT-5.6 announcement
• OpenAI: We found solutions to 10 Erdős problems
• DeepSeek: API pricing
• apito.ai: Homepage

Disclosure

ClaudeAPI / apito.ai is an independent third-party technical service and is not affiliated with OpenAI, DeepSeek, or Anthropic. Pricing, model availability, quotas, and routing behavior can change; always verify the live provider console before production rollout.