Claude Opus 5.5 Complete Guide: Fable 5.1-Level Performance for Less Than Opus 5 — Pricing, Plans, Migration

Anthropic released Claude Opus 5.5 on September 22, 2026. It is the first model in the new "Claude 5.5 family", and according to the official announcement it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5 on typical workloads. API pricing is $4 per million input tokens and $20 per million output tokens, 20% below Opus 5 ($5/$25), and it is available on the paid plans (Pro, Max, Team, Enterprise). This guide, based on the official model docs and announcement, covers what changed, pricing and plan terms, and what you need to fix when moving API code from Opus 5.

What changed

The headline is Fable 5.1-class performance at a lower cost than Opus 5. The main points from Anthropic's September 22, 2026 announcement:

• Performance — described as "at the level of Claude Fable 5.1 on most work". The announcement's comparison figures are Terminal-Bench 4.0 66.4% (Opus 5: 52.3%), FrontierCode 54.4% (48.0%), CursorBench 57.8% (46.6%) and GDPval-AA v2.1 1846 Elo (1708). These are Anthropic's own published numbers and depend on evaluation setup.
• Cost — "40% less to run than Opus 5 on typical workloads" at default settings. This likely combines the 20% price cut with the lower default effort and the 60% cheaper cache reads.
• Speed — output generation is "more than 30% faster than Opus 5". Fast mode (the same model on a faster processing path) is available in Claude Code and the Claude Platform at up to 2.5x speed; on the API it is priced at $8 input / $40 output.
• Vision — per the docs' behavioral notes, it reads charts, diagrams and screenshots more accurately without tools.
• Safeguards — a biology classifier and a reasoning_extraction category were added. Such requests come back with stop_reason: "refusal" and a category in stop_details (cyber, bio, reasoning_extraction and others).

The announcement also says "Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks". No dates have been set officially.

Pricing and specs

API pricing is as follows (official model docs, checked September 27, 2026). All figures are US dollars per million tokens.

Opus 5 remains available, but the official model docs label it "Active (legacy)" and show a banner recommending migration to Opus 5.5. Compared with Fable 5.1 ($10/$50), Opus 5.5 costs 40% as much on both input and output.

How it works on plans

In short, it is available on the paid Pro, Max, Team and Enterprise plans and not on Free. The official pricing page (claude.com/pricing, checked September 27, 2026) lists Opus as available on Pro, Max, Team and Enterprise and not on Free. Fable has separate terms (included up to 50% of the weekly limit on Max and Team Premium), covered in our Fable 5.1 guide.

Two usage-policy changes came with the same announcement.

• Higher five-hour limits — Anthropic said it is "increasing five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans". The size of the increase is not stated in official materials.
• A usage reset — "We're also providing subscription users a rate limit reset, which you can now save and use whenever you choose." The number of resets and their expiry are not specified officially (some reports say the in-app reset expires on October 22, but there is no official confirmation).

Whether the default model in the app or Claude Code has switched to Opus 5.5 is not confirmed in official documentation. The reliable way to know which model you are on is the model picker in the app (see changing model and effort settings in the Claude app).

What to fix when moving API code from Opus 5 to 5.5

Often only the model ID changes, but the official "What's new" page lists five breaking changes. If any applies to you, requests are rejected with a 400 error or the response shape changes. This section is for people writing code against the API, so it is heavy on jargon; if you only use the app or Claude Code, feel free to skip it.

Terms first

What the code change looks like

Three common cases, before and after. All use the Python SDK; field names are the same in other languages.

① Code that turned thinking off → control it with effort

# Opus 5 (now a 400 error)
client.messages.create(
    model="claude-opus-5",
    thinking={"type": "disabled"},          # ← the problem line
    ...
)

# Opus 5.5
client.messages.create(
    model="claude-opus-5-5",
    output_config={"effort": "low"},        # low when speed matters
    ...
)

② Code that forced a tool → strict + auto

# Opus 5 (now a 400 error)
tool_choice={"type": "tool", "name": "get_weather"}

# Opus 5.5
tools=[{
    "name": "get_weather",
    "description": "Current weather for a city name",
    "input_schema": {...},
    "strict": True,                         # guarantees the input shape
}],
tool_choice={"type": "auto"}               # the model decides whether to call it

③ Pick the answer by block type, not position

# Risky: if the first block is thinking, this is not text
answer = response.content[0].text

# Safe: join only the blocks whose type is "text"
answer = "".join(
    block.text for block in response.content if block.type == "text"
)

Other things worth knowing:

• Default effort went from high to medium — the same code may think less deeply. The official migration guide recommends setting effort explicitly and re-baselining cost and latency at the chosen level; at xhigh or max it suggests starting max_tokens (maximum reply length) at 64K.
• The first content block may be thinking — select blocks by their type field rather than by position such as content[0].text (example ③ above).
• Handle refusals — process stop_reason: "refusal" (a safety refusal) with its stop_details category (cyber, bio and so on — the reason), and configure server-side fallback (routing to another model on refusal) if needed.
• Unchanged — rejection of non-default sampling parameters (temperature, top_p, top_k, the randomness controls) and of assistant prefill was already the case on Opus 5, so nothing changes there. The 1M context window (up to one million tokens of input per request) is the default with no beta header.
• New beta features — defining tools in a message (inline-tools-2026-09-15), compaction on demand (compact-2026-09-04, summarising a long conversation to save tokens), per-message effort (a different thinking depth per turn), mid-conversation system messages and task budgets (a token ceiling per task) are supported. Betas are enabled by naming them in the anthropic-beta header.

Put together, the minimum request shape after migrating looks like this:

client.messages.create(
    model="claude-opus-5-5",
    max_tokens=16000,
    output_config={"effort": "medium"},   # thinking is always on; effort is the only control
    tools=[{**tool, "strict": True} for tool in tools],
    tool_choice={"type": "auto"},
    messages=[{"role": "user", "content": "..."}],
)

How to Use It

• Everyday users — on Pro or above, pick Opus 5.5 in the model selector. Anthropic's official recommendation is to "start with Claude Opus 5.5 for most workloads" and use Fable 5.1 "for demanding reasoning and long-horizon agentic work, or when your evals on Claude Opus 5.5 at higher effort still fall short".
• API users — before changing the model ID in Opus 5 code, run through the five checks above, set effort explicitly and re-measure cost. With the minimum cacheable prompt down to 512 tokens, even short system prompts qualify for caching. Estimate costs with our API cost calculator.
• Claude Code users — Fast mode is available in Claude Code too. Turn it on for speed-sensitive repetitive work and off for precision work.

Frequently asked questions

Q. Can I still use Opus 5?
Yes. The official model docs mark Opus 5 as "Active (legacy)" with retirement not sooner than July 24, 2027, while recommending migration to 5.5.

Q. Should I use Opus 5.5 instead of Fable 5.1?
Anthropic says Opus 5.5 is enough for "most workloads" and recommends Fable 5.1 for demanding reasoning and long-horizon agentic work. Opus 5.5 costs 40% of Fable 5.1's price.

Q. Is it available on the Free plan?
No. The official pricing page lists Opus on Pro, Max, Team and Enterprise, and not on Free.

Q. How do I get the usage reset?
It was announced for subscription users, but the number of resets, expiry and exact steps are not detailed in official materials. The usage screen in the app is the most reliable place to check.

Related articles

• Claude Opus 5 complete guide
• Claude Fable 5.1 complete guide
• Claude model selection guide
• Claude API pricing by model
• Claude usage limits explained

Sources: Anthropic "Introducing Claude Opus 5.5" (2026-09-22); Claude Platform docs "Claude Opus 5.5", "What's new in Claude Opus 5.5", "Migrating to Claude Opus 5.5", "Claude Opus 5" and the model deprecations page; claude.com/pricing (all checked September 27, 2026).