Claude in 2026: Pricing, Opus 5 and Fable 5.1, and What Happened to Claude Code's Limits on September 14

Claude's pricing page has not been redesigned in a while: four tiers, a list of what each one unlocks. What is worth catching up on is what sits underneath that page right now - a four-model lineup that does not work like a version ladder, and a change to Claude Code's usage limits that took effect on September 14, 2026, and quietly made the most expensive tier feel smaller than it did all summer. Here is a plan-by-plan breakdown, what the model names actually mean, and what the September 14 change does and does not affect.

The Claude plans, tier by tier

There is one free plan and four paid tiers, plus team and enterprise pricing on top. The jump between tiers is mostly about how much you can use in a session and which models you can reach, not a completely different product at each step.

• Free - $0. Runs Sonnet 5 as the default model, with a 200K-token context window. Includes chat on web, phone and desktop, web search, memory, file creation, code execution and voice mode.
• Pro - $20/month, or $17/month billed annually. Adds access to Opus 5, a larger session allowance than Free, Claude Code, and Google Workspace integration.
• Max 5x - $100/month. Five times Pro's session limit, same model access as Pro.
• Max 20x - $200/month. Twenty times Pro's session limit, and the only tier where Fable 5.1 is available at all.
• Team Standard - $25/seat/month ($20/seat billed annually) and Team Premium - $125/seat/month ($100/seat billed annually), both for teams of 2 to 150.
• Enterprise - from $20/seat/year, plus usage billed at API rates on top.

One detail that trips people up: Free and Pro give you the same base model, Sonnet 5, by default. The difference between those two tiers is mostly about how much you can use in a session and which extra tools you get, not which model answers you - Opus 5 only becomes reachable once you are on Pro or above, and Fable 5.1 only on Max. Claude Code itself is a separate agentic coding tool available from Pro upward, and it runs on its own weekly usage limits, entirely separate from the limits on regular chat.

Fable 5.1, Opus 5, Sonnet 5, Haiku 4.5: four models, not four versions

The names read like a ladder, and that reading is wrong. These are four different models built for different jobs within the same generation, not sequential releases where each one replaces the last - the same pattern you will find with most other frontier labs right now, just under different names.

• Fable 5.1 is the most capable model Anthropic has released to the general public, built for the heaviest reasoning, the longest agentic runs, complex code, research, and work across documents, spreadsheets and presentations. It is only reachable on Max.
• Opus 5 is the flagship for agentic coding and enterprise work, with a 1M-token context window and up to 128K tokens of output. It becomes available starting on Pro.
• Sonnet 5 is built for agentic code, tool use and reasoning at a lower cost than Opus 5 - it is the default model on Free and Pro.
• Haiku 4.5 is the fastest and cheapest of the four, meant for quick, high-volume tasks rather than the hardest reasoning problems.

Picking a plan, in practice, is picking which of these four you actually need reachable. Someone doing everyday chat and writing rarely needs to leave Sonnet 5 on Free or Pro. Someone running Claude Code against a real codebase usually wants Opus 5 at minimum, which means Pro or above. Fable 5.1's long-horizon reasoning is the one capability that is genuinely gated behind the $200 tier, and it is also the model most directly involved in the limits story below.

What happened to Claude Code's limits on September 14

This is the part of the pricing story that a plain feature comparison will not show you, because it is not about what a tier unlocks - it is about how far the same tier now stretches.

Since May 2026, Claude Code had been running a temporary 50% increase on top of its base weekly usage limits. That temporary boost ended on September 13, 2026, and a new, permanent increase took effect the next day, on September 14 - but the permanent bump is only 25% over the original base limit, not 50%.

Do the arithmetic on that and the practical effect is a step down, not up, for anyone who had gotten used to the May boost: roughly 17% less weekly capacity than they had access to for the past several months. Anthropic's own framing is that it made the extra headroom permanent while it works on giving users better visibility and control over their usage - a real improvement in stability, just a smaller number than the one people had adapted their workflows around.

The tier that absorbed the most complaints is Max at $200/month. Subscribers reported burning through their entire weekly Claude Code allowance in a single task or a short session, a problem that gets worse fast when the work is running on Fable 5.1, which is both the most capable and the most token-hungry model in the lineup. Some of those subscribers cancelled outright; others moved parts of their workload to competing coding tools rather than wait out the tighter ceiling.

If your Claude Code usage sits comfortably under the new permanent limit, none of this changes anything for you day to day. If you were running close to the May-era ceiling, the honest read is that the room you had gotten used to is gone, and Anthropic has not signaled a return to the temporary 50% level.

Where a card actually gets declined, and where it does not

It is worth being precise here, because Claude is not a case where a card simply fails to clear locally the way it does with some other AI subscriptions. In places where a bank-issued card is declined at checkout for routing reasons that have nothing to do with the product itself - a pattern that shows up on plenty of unrelated US-billed subscriptions in India, Brazil, Turkey, Nigeria and Argentina - the fix is usually just a different way to pay the same service. That is the specific, narrow problem a local billing route solves: a card that will not clear, for a service that is officially available where you are.

We route AI subscriptions for client work through SYNTX for exactly that reason: one subscription and one card instead of four or five separate ones, billed through a route that actually clears. Check the catalog before you pay, since what is listed there shifts over time, and confirm whichever model you need is actually on it before committing.

Frequently asked questions