GPT-6 Astra Price: ChatGPT Plans ($100, $200 Pro) and API Cost (2026)

The launch post covers what GPT-6 Astra is. This one covers what it costs, in real dollars, with the tiers OpenAI buried in the fine print.

Update, October 7, 2026: OpenAI released GPT-6 Sol at $2 / $10 and GPT-6 Luna at $0.10 / $0.50 per million tokens on September 22, half the price of the GPT-5.6 tiers they replace. Astra is now 5x Sol, and the comparisons below use the new rates. The GPT-6 Sol and Luna pricing post has the full rate card and how the open models price against it.

GPT-6 Astra shipped on September 3, 2026 at double the previous flagship's list rate, and with GPT-6 Sol now at $2 in and $10 out, it costs 5x the tier below it. The headline numbers are $10 per million input tokens and $50 per million output. If you're asking what Astra costs in ChatGPT rather than the API, the short answer is nothing extra: it's included as GPT-6 Pro on the Pro, Business, and Enterprise plans with a weekly or monthly message cap, and Plus gets it inside ChatGPT Work and Codex; the full plan table is at the end of this post. But there are six more numbers on OpenAI’s pricing page that decide what you’ll actually pay: cached input, cache writes, long-context rates, batch, flex, and Fast mode. Here’s all of it, with the math done.

The full price list

Four things that matter more than they look.

Cached input is 90% off. If your application reuses a long system prompt, tool definitions, or a document across calls, the repeated portion bills at $1 instead of $10. For agent workloads, which resend most of the context every turn, this is the single biggest lever on the bill. Cache writes cost $12.50 per million, so the first pass is more expensive than the list price and every reuse after it is far cheaper.

Long context has its own rate card. Requests with more than 272K input tokens bill the whole request at $20 in and $75 out. The 1,050,000-token window is real, but using the back half of it costs double. Plan for that before you design around the full window.

Fast mode is a straight 2x for up to 2x speed, and OpenAI notes it’s not available for GPT-6 Astra with EU data residency.

Batch and flex are half price. Anything that can wait, evaluations, backfills, offline processing, should go through one of those and pay $5 in and $25 out.

Against GPT-6 Sol

GPT-6 Sol replaced GPT-5.6 Sol on September 22 at half the price, which makes Astra 5x Sol on input, cached input, and output. At launch the ratio was 2.5x against GPT-5.6 Sol's $4 in and $20 out.

The bottom of the family for scale: GPT-6 Luna is $0.10 in / $0.50 out. Astra costs 100x Luna on both sides. That's the spread OpenAI now expects you to route across.

What it costs in practice

Take a mid-sized agent deployment: 10 million input tokens and 2 million output tokens a day, with a 70% cache-hit rate on input, which is typical once system prompts and tools are stable.

Two lessons in that table. Caching cuts Astra’s input bill by nearly two-thirds, but output is the cost that doesn’t move, and Astra’s output rate is the expensive one. And GPT-6 Sol does the same volume for 20% of the price. Whether Astra earns the difference depends entirely on whether your workload is in the tier where its gains show up: terminal tasks, computer use, long-horizon engineering, and cyber-adjacent work, per the launch benchmarks.

Against the open field

The other comparison worth running is against the open-weight models most production teams already route to. Vendor-direct API rates, as of early September 2026:

Astra's output token costs 8x Qwen 3.8-Max's, 40x DeepSeek V4.1 Flash's, and 100x GLM 5.3 Flash's. Those are different capability classes and the comparison isn’t “which is better.” It’s “which traffic needs the $50 model.” For most teams the answer is a small slice: the hardest agentic and engineering tasks. The everyday middle runs on the open models at a tenth of the cost or less, and a routing layer sends each request where it belongs. How to switch models without changing your code covers the mechanics; the best OpenAI API alternatives covers the field. Meta’s answer at the mid-tier is covered in the Muse Wonder pricing post: $4 in and $20 out, 60% under Astra on both.

That’s what Yotta AI Gateway is for: Qwen 3.8-Max, Qwen3.8-27B, DeepSeek V4 Pro and V4 Flash, GLM 5.3, and Kimi K3 behind one API key with OpenAI-compatible endpoints and flat pricing, with Claude being added. Put Astra in the same routing config for the tail and the bill stops being a flagship bill.

GPT-6 Astra in ChatGPT: subscription price and limits by plan

In ChatGPT the model shows up as GPT-6 Pro, which OpenAI describes as "powered by GPT-6 Astra." There's no separate Astra subscription and no announced price change to any plan; you pay the plan price and get a weekly or monthly allowance of GPT-6 Pro messages on top of normal usage. Plus subscribers get Astra inside ChatGPT Work and Codex but won't see it in the regular chat model picker.

If you're deciding between a subscription and the API, the rule of thumb holds: a seat is cheaper for a person, the API is cheaper for a program. Two hundred GPT-6 Pro messages a week on the $200 Pro plan is a fixed cost; the same volume on the API depends entirely on how many tokens each message carries.

Frequently asked questions

How much does GPT-6 Astra cost? $10 per million input tokens and $50 per million output on the standard API tier. Cached input is $1, batch and flex are half price, Fast mode is double, and requests over 272K input tokens bill at $20 in and $75 out.

Is GPT-6 Astra more expensive than GPT-6 Sol? Yes, 5x on input and output. GPT-6 Sol is $2 in and $10 out per million tokens, and Astra is $10 and $50.

What does cached input mean for Astra? Repeated input tokens, like a stable system prompt or tool schema, bill at $1 per million instead of $10 once cached. Writing to the cache costs $12.50 per million, so caching pays off on anything reused more than a couple of times.

Is there a cheaper way to run GPT-6 Astra? Batch API and flex processing at 50% off for work that can wait, and aggressive prompt caching for work that can’t. There's no cheaper model tier of Astra itself; the cheaper option is routing lower-value traffic to GPT-6 Sol, GPT-6 Luna, or an open model.

How does GPT-6 Astra pricing compare to open-source models? Astra's $50 output rate is roughly 8x Qwen 3.8-Max, 40x DeepSeek V4.1 Flash, and 100x GLM 5.3 Flash at vendor rates. Most teams use Astra for the hardest tier and route the rest to cheaper models.

Can I use GPT-6 Astra alongside Yotta AI Gateway? Yes. Astra is called through OpenAI or AWS, and Yotta AI Gateway carries Qwen, GLM, DeepSeek, and Kimi models behind one OpenAI-compatible key, with Claude being added. A routing layer can send traffic to both.

Does GPT-6 Astra Fast mode cost more? Yes, exactly double standard rates for up to 2x speed, and it isn’t offered with EU data residency.

How much is GPT-6 Astra in ChatGPT? There's no extra charge. It's included as GPT-6 Pro on the Pro $100, Pro $200, Business, and Enterprise plans with a weekly or monthly message cap, and Plus subscribers get it inside ChatGPT Work and Codex. Free and Go plans don't include it.

How much is a GPT-6 Astra subscription per month? There's no separate Astra subscription. The lowest ChatGPT plan with GPT-6 Pro in the chat picker is Pro at $100 a month, with 50 GPT-6 Pro messages a week. Pro at $200 a month gets 200 a week.

Bottom line

GPT-6 Astra is priced as a flagship for the hardest work, not a default for everything: $10 in, $50 out, 5x GPT-6 Sol, and 8 to 100x the open models on output. Caching and batch pull the input side down a lot; nothing pulls the output side down except sending less traffic to it. The teams that come out ahead will treat Astra as one route in a multi-model setup, not the setup.

The full launch breakdown, rollout schedule, and benchmarks are in GPT-6 Astra: Release Date, Pricing, Benchmarks, and Rollout. For the open-model side of the routing table, Yotta AI Gateway is one API key away.