OpenAI released GPT-6 Astra on September 3, 2026, and it is now live in the API and rolling out across paid ChatGPT plans. Astra costs $10 per million input tokens and $50 per million output tokens, which is 2.5 times the price of GPT-5.6 Sol, and it ships with a 1,050,000-token context window. All pricing and specifications in this article were checked against OpenAI’s official API pricing and model documentation in September 2026.
What OpenAI Announced on September 3, 2026
GPT-6 Astra is OpenAI’s new flagship model, positioned as a step up for computer use, browsing, software engineering, and long agentic sessions rather than a general jump in raw reasoning scores. It went live in the API on launch day and simultaneously reached ChatGPT Pro, Enterprise, and Business Premium users inside the Work and Codex surfaces.
The rollout was deliberately staged. Pro, Enterprise, and Business Premium accounts got Astra immediately. Plus and standard Business subscribers were told access would arrive over the following days, and it did, though initially only inside ChatGPT Work and Codex rather than the default chat interface. That distinction confused a lot of subscribers in the first week, and it is the single most common reason a Plus user does not see Astra in their model picker.
Two things make this launch matter more than a routine version bump. First, the price moved up rather than down, breaking the pattern of the last several OpenAI releases. Second, the context window crossed one million tokens, which changes what kinds of jobs are practical to send in a single request.
GPT-6 Astra API Pricing and Specifications
The headline number is a 150 percent price increase over the model most developers were already running. Astra bills at $10 per million input tokens and $50 per million output tokens. Cached input drops to $1 per million, and cache writes are billed at $12.50 per million, so prompt caching is no longer a minor optimization on this model. It is the difference between a workable bill and a painful one for any application that reuses a large system prompt.
Here is how Astra compares against the current OpenAI lineup, using standard-tier list prices verified in September 2026.
| Model | Input per 1M | Cached input per 1M | Output per 1M | Relative output cost |
|---|---|---|---|---|
| GPT-6 Astra | $10.00 | $1.00 | $50.00 | 1.0x |
| GPT-5.6 Sol | $4.00 | $0.40 | $20.00 | 0.40x |
| GPT-5.6 Terra | $2.00 | $0.20 | $12.00 | 0.24x |
| GPT-5.6 Luna | $0.20 | $0.02 | $1.20 | 0.024x |
| GPT-5.4-mini | $0.75 | $0.075 | $4.50 | 0.09x |
The core specifications are equally important for planning. Astra carries a 1,050,000-token context window, accepts up to 922,000 input tokens, and can produce up to 128,000 output tokens in a single response. Its knowledge cutoff is April 30, 2026. It accepts text and image input and returns text only.
Long-Context and Fast Mode Surcharges
The list price is not the whole bill. Any request carrying more than 272,000 input tokens is billed at double the input and cache rates and 1.5 times the output rate for the entire request, not just the portion above the threshold. That is a meaningful cliff. A 280,000-token prompt costs substantially more than a 270,000-token prompt, so it is worth trimming context to stay under the line when you can.
Fast mode doubles the applicable rates in exchange for roughly twice the throughput. In the other direction, Batch and Flex processing carry a 50 percent discount, which is the single easiest lever for any workload that does not need an immediate answer. Overnight evaluation runs, bulk classification, and document processing pipelines are obvious candidates.
What Astra Does Not Support
Astra is available on the Chat Completions, Responses, and Batch endpoints. It does not support Realtime, Assistants, fine-tuning, embeddings, or audio and video processing. If your product depends on any of those, Astra is not a drop-in replacement and you will be keeping an older model in the stack regardless. On the standard rate limit ladder, Tier 5 accounts get 15,000 requests per minute and 40 million tokens per minute.
Who Actually Gets Astra in ChatGPT
Access in the consumer and business apps does not map cleanly onto access in the API, and the naming makes it worse. In ChatGPT, the highest-effort variant appears as GPT-6 Pro, which is powered by Astra, and OpenAI lists it for Pro $100, Pro $200, Business, and Enterprise plans. Plus subscribers get Astra itself, but the practical entry point has been ChatGPT Work and Codex rather than the standard chat model picker.
| Plan | Monthly price | Astra access | Notes |
|---|---|---|---|
| Free | $0 | No | Older models only |
| Plus | $20 | Yes, staged | Surfaced in Work and Codex first |
| Pro | $100 | Yes | Roughly 5x the usage allowance of Plus |
| Pro | $200 | Yes | Roughly 20x the usage allowance of Plus |
| Business | Per seat | Yes | Premium seats got it on launch day |
| Enterprise | Custom | Yes | Available at launch |
Astra usage is included inside existing subscription allowances rather than metered separately, and OpenAI has said individuals and businesses can buy additional credits when they exhaust their allowance. If you are on Plus and running long Codex sessions, the $100 Pro tier exists specifically for that gap, sitting between the standard Plus allowance and the full $200 tier.
Is Astra Worth 2.5 Times the Price of GPT-5.6 Sol
This is where the answer genuinely splits by use case, and the honest version is that Astra is not a uniform upgrade.
On general reasoning, independent third-party evaluation put Astra at the same composite intelligence score as GPT-5.6 Sol, with Anthropic’s Fable 5.1 several points ahead of both. Because Astra costs 2.5 times more per token, the same test suite runs roughly 75 percent more expensive per task at maximum effort. If your workload is summarization, classification, extraction, or general question answering, upgrading buys you a larger bill and very little else. Sol, Terra, or Luna remain the rational choices depending on how much capability you actually need.
Agentic coding is the opposite story. Astra is meaningfully more token efficient on coding work, using roughly a third of the tokens Sol needs on comparable tasks, which claws back much of the per-token premium. On terminal and agentic benchmarks the gap is large, with Astra scoring in the high fifties against Sol’s high thirties on one widely cited terminal agent evaluation. On composite coding agent scores it lands within a few points of Fable 5.1 while running at less than half the cost of the previous Anthropic flagship.
The pattern is consistent: Astra earns its price when the job involves long tool-using sessions, browsing, computer use, or multi-step software engineering. It does not earn its price on single-turn text work.
How to Decide What to Run This Month
A few concrete rules will save most teams more money than any amount of prompt tuning.
- Do not migrate by default. Route your traffic by task type. Coding agents and browser or computer-use workflows go to Astra. Everything else stays on Sol, Terra, or Luna.
- Turn on prompt caching before you turn on Astra. At a 10 to 1 ratio between standard and cached input, a stable system prompt reused across requests is the highest-leverage change available.
- Watch the 272,000-token line. Trimming a prompt from 280,000 to 265,000 tokens does not shave 5 percent off the cost, it removes a multiplier applied to the whole request.
- Move anything asynchronous to Batch or Flex. The 50 percent discount applies to the full request, and most evaluation and bulk-processing work has no latency requirement.
- Measure tokens, not just token price. Astra’s efficiency gain on coding tasks means the per-task cost comparison can favor it even though the per-token price does not. Run your own workload before deciding.
- Keep a fallback model wired up. Astra does not support Realtime, fine-tuning, or embeddings, so a mixed-model architecture is not optional if you use those features.
One practical caution: model pricing and free tier allowances in this market have changed within weeks of a launch more than once this year. Everything above reflects OpenAI’s published rates as of September 2026, and it is worth re-checking the official pricing page before you commit to a budget for the quarter.
GPT-6 Astra FAQ
Is GPT-6 Astra included in ChatGPT Plus at $20 a month?
Yes. Astra reached Plus subscribers in the days following the September 3, 2026 launch, though the initial access point was ChatGPT Work and Codex rather than the default chat model picker. The separate GPT-6 Pro variant, which runs Astra at higher effort, is listed for Pro $100, Pro $200, Business, and Enterprise plans rather than standard Plus. Usage counts against your existing plan allowance.
How much does GPT-6 Astra cost compared to GPT-5.6 Sol?
Astra is 2.5 times more expensive per token. Astra bills $10 per million input tokens and $50 per million output tokens, while GPT-5.6 Sol bills $4 and $20 respectively. Cached input is $1 per million on Astra versus $0.40 on Sol. Per completed task the gap narrows on coding work, where Astra uses roughly a third of the tokens Sol needs, but it widens on general reasoning where the two models score about the same.
What is the real usable context window on GPT-6 Astra?
The context window is 1,050,000 tokens, with a maximum of 922,000 input tokens and 128,000 output tokens per request. The important caveat is the pricing threshold: any request exceeding 272,000 input tokens is billed at double the input and cache rates and 1.5 times the output rate across the entire request. The full million-token window is technically available, but it is a premium feature in practice rather than a free capacity increase.
Clear, fact-checked guides on money, tech and everyday decisions.
Related reading
- DeepSeek Peak-Hour Pricing: What Changes on August 16, 2026
- OpenAI Just Cut GPT-5.6 Sol API Prices by Up to 33%
- Claude Fable 5.1 and Mythos 5.1: 3 Breaking API Changes Developers Must Fix
More on this topic: all AI Tools articles · full article index