Claude Mythos Preview is not the cheapest Claude API route most people can use today. Anthropic still treats Mythos as an invitation-only preview with participant-only pricing, so it is not the normal public buy path.
As of April 12, 2026, the cheapest official Claude API model you can actually use is Claude Haiku 4.5 at $1 per million input tokens and $5 per million output tokens. If your workload can tolerate asynchronous turnaround, Anthropic's Batch API is the main official discount lever because it cuts those prices in half.
The practical route choice is straightforward. Start with Haiku if you want the cheapest official Claude path, move up to Sonnet or Opus only when the capability gain matters, and treat cheaper third-party Claude access as a separate trust and routing decision rather than as proof that Mythos is the public bargain route.
Pricing and access status below reflect Anthropic's Models overview, Claude API pricing docs, and Project Glasswing page as of April 12, 2026.
What matters right now
| Question | Current answer | Why |
|---|---|---|
| Is Claude Mythos Preview the cheapest Claude API route I can buy today? | No | Anthropic still treats Mythos as an invitation-only preview rather than a public self-serve route. |
| What is the cheapest official Claude API route? | Claude Haiku 4.5 | Anthropic's public pricing table lists Haiku 4.5 at $1 / MTok input and $5 / MTok output. |
| What is the main official discount lever? | Batch API | Anthropic says Batch API cuts supported model pricing by 50%. |
| Does Mythos have any published pricing at all? | Yes, but only inside the preview boundary | Anthropic's Project Glasswing page lists participant pricing of $25 / MTok input and $125 / MTok output after credits. |
| What if another provider sells Claude cheaper than Anthropic direct? | Treat that as a separate decision layer | A cheaper gateway or provider may help, but that is not the same thing as Mythos being a public cheap route. |
What Mythos pricing actually means
What confuses buyers is that Anthropic has published a real Mythos price without turning Mythos into a public buy-now model.
The first truth is that Anthropic does publish real Mythos pricing now. On the current Project Glasswing page, Anthropic says Mythos Preview participant pricing is $25 per million input tokens and $125 per million output tokens after the initial credit period. That is much more concrete than the rumor-driven pages that still talk as if Mythos has no public pricing surface at all.
The second truth is the one that matters more for procurement. Anthropic does not present that pricing as a normal public buy-now route. The same Project Glasswing page frames Mythos Preview as an invitation-only frontier model for defensive cybersecurity workflows. Anthropic's Models overview also says there is no self-serve sign-up for Mythos Preview. That means participant pricing is real, but it is still participant pricing inside a gated preview contract.
That distinction matters more than the price row itself.
If you are already a Project Glasswing partner or one of the invited organizations Anthropic mentions, Mythos pricing is operationally relevant. If you are an ordinary reader trying to decide what to deploy this week, it is not yet the answer to your buying question. Treating participant pricing as if it were normal public pricing is exactly how cheapest Claude API pages end up sounding precise while still giving the wrong route.
The clean mental model is this:
- Mythos Preview pricing tells you what the preview audience pays.
- Anthropic's public API pricing table tells you what ordinary self-serve buyers can actually choose today.
- Those are related surfaces, but they are not the same contract.
Once you sort that out, the rest of the article becomes much simpler.
The cheapest official Claude API route right now
Anthropic's public Claude API pricing page currently lists three broadly available price rows:
| Model | Direct input price | Direct output price | Batch API price | Practical role |
|---|---|---|---|---|
| Claude Haiku 4.5 | $1 / MTok | $5 / MTok | $0.50 / MTok input, $2.50 / MTok output | Cheapest official route when cost matters most |
| Claude Sonnet 4.6 | $3 / MTok | $15 / MTok | $1.50 / MTok input, $7.50 / MTok output | Mainstream route when you need more capability |
| Claude Opus 4.6 | $5 / MTok | $25 / MTok | $2.50 / MTok input, $12.50 / MTok output | Premium route when output quality matters more than cost |
That table is why the direct answer is not very controversial once Mythos is removed from the wrong branch. Haiku 4.5 is the cheapest official Claude API route ordinary users can actually use today.
There are two important consequences.
First, the cheapest official answer is about the public Claude lineup, not about Anthropic's most interesting preview project. People often blur those together because both belong to Anthropic's broader model story. Your budget does not care about that narrative convenience. It cares whether the route is public and what the current price row is.
Second, Batch API matters more than many quick comparison posts admit. If your workload is asynchronous and you do not need instant responses, Anthropic's own Batch discount can cut the official Claude price floor in half. That is the first official lever to check before you jump to third-party pricing boards that mix direct models, gateways, credits, and promotional balances into one chart.
This is also where readers often overreact to the headline word cheapest. The cheapest official route is not automatically the best route for every real workload. Haiku is the official floor. It is the right starting point when budget pressure is high, throughput matters, or you are validating whether Claude is affordable at all. It may not stay the right answer if your task quality requirements quickly force you onto Sonnet or Opus.
So the useful decision rule is:
- Start with Haiku 4.5 if your question is genuinely about the cheapest official Claude path.
- Move to Sonnet 4.6 only when the quality gain justifies the higher price.
- Move to Opus 4.6 only when you already know the premium capability is worth paying for.
- Check Batch API before you assume Anthropic direct is too expensive.
That is a much better buying sequence than chasing Mythos headlines first.
When Haiku is the right answer and when it isn't
The phrase cheapest Claude API hides a second decision that matters just as much as the price table: what kind of work you are trying to pay for.
Haiku is the right answer when you need the official price floor. That usually means cost-sensitive automation, first-pass classification, lightweight extraction, draft summarization, or early-stage product validation where the question is still Can I make this economics work at all?
Haiku stops being the whole answer when the failure cost of a weaker model becomes more expensive than the token savings.
If your workflow depends on stronger reasoning, higher coding reliability, or fewer retries, Sonnet can become the cheaper practical route even though it is not the cheapest line item. The same logic applies to Opus for narrower high-stakes tasks. This is why honest price guidance should always separate cheapest official price row from lowest total workflow cost. Anthropic's table answers the first question. Your workload answers the second.
That distinction also protects the article from sliding into the opposite mistake. Some comparison pages correctly notice that Haiku is the official floor, then overcorrect by pretending the cheapest route discussion ends there. It does not. The official floor is the baseline. The real route decision is still about what level of capability your workload needs and whether Anthropic's own Batch discount already gets you low enough.
If your next question is not What is the cheapest official route? but instead How much Claude capability do I get at each tier? then you are leaving the price-floor problem and entering a different reader job. That is exactly why Mythos headlines and official public pricing should not be collapsed into one page promise.
What about cheaper third-party Claude access?
Yes, a third-party provider or gateway can be cheaper than Anthropic direct for some Claude models on some days. That can be a legitimate route if your real problem is regional payment friction, a bundled model catalog, a temporary credit promotion, or a lower public price on Sonnet or Opus than Anthropic currently lists.
But that is a different answer from the official-baseline question above.
Once you leave Anthropic direct, you are no longer asking only about model price. You are also asking about routing quality, stability, logging, regional availability, support, and whether the provider's public price board still matches the real bill. That can still be worth it. It just should not be smuggled into the page as if it were the same kind of truth as Anthropic's own pricing table.
The clean ordering is:
- Decide the official baseline first.
- Decide whether Anthropic direct already works for your budget, especially with Batch API.
- Only then compare whether a third-party route improves cost, access, or operational convenience enough to justify the extra trust layer.
There is no stable universal cheapest Claude provider overall answer. That answer changes quickly and depends on model, region, credits, and routing terms. What does not change as quickly is the official baseline: Haiku 4.5 is Anthropic's cheapest public Claude API route on April 12, 2026, and Mythos Preview is still not a public self-serve answer for ordinary users.
If your real question is adjacent to this one
This article is most useful when your question is narrow:
“"I saw Mythos pricing or Mythos headlines. Can I actually buy that route, and if not, what is the cheapest Claude API path I can use today?"
If your real question is different, the better next step is usually a sibling page.
If you really want to know whether Claude has any ordinary public free route or starter-credit path, read Claude API key free tier. That page is built for the what is free, what is starter credit, what becomes paid question instead of the Mythos availability confusion.
If your real question is whether Mythos should change what model you use now, read Claude Mythos Preview vs GPT-5.4. That page focuses on the use now versus track later decision rather than the cheapest-route question.
If your real question is broader unreleased-Claude naming confusion, read Claude Capybara. That page is the better place to sort Anthropic's unreleased-model naming and public-route boundaries without overloading this pricing guide.
Those sibling routes keep the pricing answer narrow enough to stay useful instead of turning it into a low-quality Claude roundup.
FAQ
Does Anthropic publish Mythos pricing right now?
Yes. Anthropic's current Project Glasswing page publishes Mythos Preview participant pricing of $25 / MTok input and $125 / MTok output after credits. The important qualifier is that this is still participant-only preview pricing, not a normal public self-serve route.
What is the cheapest official Claude API model today?
As of April 12, 2026, Anthropic's public pricing page lists Claude Haiku 4.5 as the cheapest official Claude API model at $1 / MTok input and $5 / MTok output.
Is Batch API the cheapest official Claude route?
Batch API is not a separate model. It is Anthropic's main official discount mechanism for supported models. If your workload can wait for asynchronous processing, Batch pricing cuts the listed model prices in half, which makes Haiku even cheaper.
Does Mythos make Haiku irrelevant if I only care about price?
No. Mythos participant pricing is much higher than Haiku's public pricing, and Mythos remains behind an invitation-only boundary. Even before access is considered, Mythos is not the cheap official route.
When should I consider a cheaper third-party Claude provider?
Consider it only after you know the official baseline and after you decide whether Anthropic direct plus Batch API already works for you. Third-party routes can help, but they introduce a separate trust, support, and routing layer that should not be treated as interchangeable with Anthropic's own pricing contract.
The practical takeaway
At first glance this topic looks like one pricing question, but in 2026 it really breaks into two decisions:
- Is Mythos Preview a public route I can buy now?
- If not, what is the cheapest Claude route Anthropic actually offers me today?
The first answer is no for ordinary users. The second answer is Claude Haiku 4.5, with Anthropic's Batch API as the main official way to push that cost lower.
If you remember one rule, make it this:
Mythos pricing belongs to the preview audience. Haiku pricing belongs to the public buying decision.
That single split removes most of the confusion and gives you a clean next move.