TL;DR
Last updated November 19, 2026.
Claude Opus 5.5 shipped September 22, 2026 as claude-opus-5-5. It costs $4 per million input tokens and $20 per million output, 20% less than Opus 5. Context is 1M tokens with 128K max output and a June 2026 knowledge cutoff. Anthropic reports 66.4% on Terminal-Bench 4.0, and on the public Terminal-Bench 4.1 board Claude Code with Opus 5.5 ties Fable 5.2 at 65.8% for less than a third of the run cost.
“It costs less per token than Opus 5 and uses fewer tokens per task, which nets out to a 40% drop in costs.”
What it is
Anthropic's Opus-tier model for long-running agentic coding and knowledge work. 1M context, 128K max output, adaptive thinking always on, default effort medium.
What changed from Opus 5
Price drops from $5/$25 to $4/$20 per million tokens. Terminal-Bench 4.0 rises from 52.3% to 66.4% on Anthropic's runs. Thinking can no longer be disabled, and forced tool use now returns an error.
Opus 5.5 Pricing: Tokens, Caching, and Fast Mode
Opus 4.6, 4.7, 4.8, and 5 all cost $5 per million input tokens and $25 per million output. Opus 5.5 is the first to change that price. Input drops from $5 to $4 per million tokens and output from $25 to $20. Cache reads fall from $0.50 to $0.20 per million.
| Item | Opus 5.5 | Opus 5 | Fable 5.2 |
|---|---|---|---|
| Input | $4 | $5 | $8 |
| Output | $20 | $25 | $40 |
| Cache write (5 min) | $5 | $6.25 | $10 |
| Cache write (1 hour) | $8 | $10 | $16 |
| Cache read | $0.20 | $0.50 | $0.20 |
| Fast mode (input / output) | $8 / $40 | Research preview | Available |
Fast mode runs Opus 5.5 at up to 2.5x the output speed for $8 input and $40 output per million tokens. It is available in Claude Code, on the Claude API, and on Claude Models on Google Cloud (Anthropic managed). See Anthropic API pricing for every Claude model on one table.
Opus 5.5 Benchmarks Anthropic Published
Anthropic ran every Opus 5.5 result at max effort with adaptive thinking, except Terminal-Bench 4.0, which it reports at xhigh. The GPT-6 Astra and GPT-5.6 Sol figures are OpenAI's own.
| Benchmark | Opus 5.5 | Fable 5.1 | Opus 5 | GPT-6 Astra |
|---|---|---|---|---|
| Terminal-Bench 4.0 | 66.4% | 55.8% | 52.3% | 57.9% |
| FrontierCode v1.1 (Main) | 54.4% | 50.3% | 48.0% | 53.3% |
| CursorBench 4.0 | 57.8% | 51.8% | 46.6% | not reported |
| GDPval-AA v2.1 (Elo) | 1846 | 1735 | 1708 | 1542 |
| AutomationBench | 40.0% | 31.4% | 26.9% | 41.4% |
| Humanity's Last Exam (with tools) | 67.7% | 65.6% | 63.6% | 57.2% |
| Terminal-Bench-Science 0.1 | 58.7% | 52.6% | 29.0% | 64.6% |
| OSWorld 2.1 | 81.8% | 80.7% | 74.0% | not reported |
Anthropic adds its own caveat: at this level, benchmark margins are a weaker guide to real-world differences, and in its own use the gap between Opus 5.5 and Fable 5.1 is narrower than the scores suggest. Two worked examples from the announcement carry more signal. An early tester audited and fixed a 200,000-line codebase in under three hours with Opus 5.5, where Opus 5 took over 20 hours and 2.5x the tokens. Translating HAProxy from C to Rust took Opus 5.5 9.5 hours against 12 for Fable 5.1, at 51% lower cost.
Opus 5.5 on the Terminal-Bench 4.1 Leaderboard
Terminal-Bench 4.1 replaced 4.0 on tbench.ai on November 9, 2026, with task verifiers rewritten for reward-hack resistance. It scores the agent and model together, so the Opus 5.5 entry is Claude Code running Opus 5.5.
| Agent / Model | Resolution rate | Run cost |
|---|---|---|
| GitHub Copilot / Copilot Fusion (max) | 71.2% ± 2.9 | $2.0k |
| Claude Code / Opus 5.5 (max) | 65.8% ± 3.1 | $4.7k |
| Claude Code / Fable 5.2 (max) | 65.8% ± 2.8 | $16.5k |
| Antigravity CLI / Gemini 4 Carbon (high) | 64.5% ± 3.2 | $1.4k |
| Claude Code / Sonnet 5.5 (max) | 63.9% ± 3.0 | $7.3k |
| Codex / GPT-6.2 Astra (max) | 62.1% ± 2.3 | $1.4k |
| Codex / GPT-6.2 Sol (max) | 61.5% ± 2.8 | $0.3k |
Opus 5.5 and Fable 5.2 land on the same score, and the Opus run cost 28% of the Fable run. For a terminal agent on the Claude side, Opus 5.5 is the default pick unless a task needs something Fable does that Opus does not. The full agent ranking is on best AI coding agents.
Opus 5.5 Context Window, Max Output, and Knowledge Cutoff
| Spec | Value |
|---|---|
| Context window | 1M tokens |
| Max output | 128K tokens |
| Reliable knowledge cutoff | June 2026 |
| Training data cutoff | June 2026 |
| Thinking | Adaptive, always on |
| Default effort | medium |
| Retirement | Not sooner than September 22, 2027 |
The default effort is medium, one step below Fable 5.2's high. Raise it with the effort parameter for long agent runs, and lower it where you used to disable thinking on Opus 5. See Claude context window for every model's limits.
Opus 5.5 vs Opus 5: Price, Scores, and Token Use
Opus 5.5 beats Opus 5 on every benchmark in Anthropic's table and costs 20% less per token. Anthropic says it also spends fewer tokens per task, for a 40% net cost drop. At default effort on Terminal-Bench 4.0, Opus 5.5 beats Opus 5 at max effort for about a fifth of the cost.
| Opus 5.5 | Opus 5 | |
|---|---|---|
| Price (input / output per 1M) | $4 / $20 | $5 / $25 |
| Terminal-Bench 4.0 | 66.4% | 52.3% |
| FrontierCode v1.1 | 54.4% | 48.0% |
| CursorBench 4.0 | 57.8% | 46.6% |
| Thinking off | Not allowed (400 error) | Allowed at effort high or below |
| Forced tool use | Not supported (400 error) | Supported |
Opus 5.5 Breaking API Changes and Migration
Anthropic lists four changes that can break code running on Opus 5. Thinking cannot be disabled, so thinking: {"type": "disabled"} or a manual budget_tokens returns a 400. Forced tool use is gone, so tool_choice of any or a named tool returns a 400; keep auto and use strict tool use for schema-valid JSON. Thinking blocks are tied to the model that produced them, which matters when a conversation switches models. Some older computer use tool versions are not supported.
A fifth change alters the response shape without failing a request. Text between tool calls now arrives inside thinking blocks, which are empty at the default display setting. An app that streams that text as progress goes quiet between tool calls until it sets a display value that returns the text.
Opus 5.5 API Access and Model IDs
| Platform | Model ID |
|---|---|
| Claude API | claude-opus-5-5 |
| Amazon Bedrock | anthropic.claude-opus-5-5 |
| Claude Models on Google Cloud | claude-opus-5-5 |
| Microsoft Foundry | claude-opus-5-5 |
| Claude Platform on AWS | claude-opus-5-5 |
import anthropic
client = anthropic.Anthropic()
msg = client.messages.create(
model="claude-opus-5-5",
max_tokens=4096,
output_config={"effort": "high"},
messages=[{"role": "user", "content": "Audit this module for race conditions."}],
)
print(msg.content[-1].text)Claude Opus 5.5: Pros and Cons
- $4 / $20 per million tokens, 20% below Opus 5
- 66.4% on Terminal-Bench 4.0 in Anthropic's runs, up from 52.3% on Opus 5
- Ties Fable 5.2 at 65.8% on Terminal-Bench 4.1 for 28% of the run cost
- 1M-token context, 128K max output, June 2026 knowledge cutoff
- Cache reads at $0.20 per million, down from $0.50
- Thinking cannot be disabled, and forced tool use returns a 400
- Copilot Fusion and Gemini 4 Carbon sit at or above it on Terminal-Bench 4.1
- $20 per million output is about 3.3x GLM-5.3's $6 on Morph
- Text between tool calls moves into thinking blocks, which can silence streaming UIs
Open-Weight Alternatives to Opus 5.5 on Morph
Morph does not serve Anthropic models. It serves open-weight coding models on custom kernels through an OpenAI-compatible API. GLM-5.3 runs as morph-glm53-744b with the same 1M context as Opus 5.5 at under a third of its output price.
| Model | Input | Output | Context |
|---|---|---|---|
| Claude Opus 5.5 (Anthropic) | $4.00 | $20.00 | 1M |
| GLM-5.3 (morph-glm53-744b) | $1.19 | $6.00 | 1M |
| MiniMax M3 (morph-minimax3-428b) | $0.255 | $1.02 | 256K |
from openai import OpenAI
client = OpenAI(
base_url="https://api.morphllm.com/v1",
api_key="YOUR_MORPH_API_KEY",
)
resp = client.chat.completions.create(
model="morph-glm53-744b",
messages=[{"role": "user", "content": "Explain this stack trace and propose a fix."}],
)
print(resp.choices[0].message.content)On Terminal-Bench 4.1, Claude Code with Opus 5.5 scores 65.8%; the best open-weight entry, Claude Code with Kimi K3.1 Pro, scores 59.7%, and GLM-5.3 scores 41.2%. The open-weight models cost a fraction per token. See GLM-5.2, Claude Opus 4.8, and best AI model for coding for the cross-lab comparison.
Frequently Asked Questions
What is Claude Opus 5.5?
Claude Opus 5.5 is Anthropic's Opus-tier model, released September 22, 2026, with the Claude API ID claude-opus-5-5. Anthropic built it for long-running agentic coding and knowledge work. It has a 1M-token context window, 128K max output, a June 2026 knowledge cutoff, and adaptive thinking that cannot be turned off.
How much does Claude Opus 5.5 cost?
$4 per million input tokens and $20 per million output tokens. Cache reads cost $0.20 per million, 5-minute cache writes $5, and 1-hour cache writes $8. Opus 5 costs $5/$25, so Opus 5.5 is 20% cheaper per token. Fast mode, with up to 2.5x output speed, costs $8 input and $40 output per million tokens.
What are the Claude Opus 5.5 benchmarks?
Anthropic reports 66.4% on Terminal-Bench 4.0 (xhigh effort), 54.4% on FrontierCode v1.1, 57.8% on CursorBench 4.0, 1846 on GDPval-AA v2.1, 81.8% on OSWorld 2.1, and 67.7% on Humanity's Last Exam with tools. On the public Terminal-Bench 4.1 leaderboard, Claude Code with Opus 5.5 at max effort scores 65.8%.
What is the Opus 5.5 context window?
1M tokens of context and 128K tokens of max output, the same limits as Opus 5, Sonnet 5.5, and Fable 5.2. The reliable knowledge cutoff and training data cutoff are both June 2026.
Is Opus 5.5 better than Opus 5?
On Anthropic's numbers, yes, and it costs less. Opus 5.5 scores 66.4% on Terminal-Bench 4.0 against 52.3% for Opus 5, and 54.4% on FrontierCode against 48.0%. Anthropic says it uses fewer tokens per task as well as costing less per token, which nets out to a 40% drop in cost. Four API changes can break code written for Opus 5, so read the migration notes first.
What is the Opus 5.5 model ID?
claude-opus-5-5 on the Claude API, Google Cloud, Microsoft Foundry, and the Claude Platform on AWS. On Amazon Bedrock it is anthropic.claude-opus-5-5. Anthropic says retirement will not come sooner than September 22, 2027.
What are cheaper alternatives to Claude Opus 5.5?
Morph does not serve Anthropic models. It serves open-weight coding models through an OpenAI-compatible API. GLM-5.3 (morph-glm53-744b) costs $1.19 per million input tokens and $6 per million output with a 1M context. MiniMax M3 (morph-minimax3-428b) costs $0.255 input and $1.02 output.
The fastest endpoints are private deployments
Morph's top speeds come from dedicated deployments, not shared public endpoints: speculators trained on your traffic, caching tuned to your workload, and volume discounts over public per-token rates. Over 1 trillion tokens per day run this way.
Cut the context an Opus-tier agent has to read
WarpGrep is an agentic code search tool. Point it at your repo so a 1M-token window fills with the right code instead of the whole tree. $0.80 per 100K tokens.
Sources
- Anthropic: Claude Opus 5.5 announcement (benchmarks, pricing, fast mode, availability)
- Anthropic docs: Models overview (context window, max output, cutoffs, model IDs, retirement)
- Anthropic docs: What's new in Claude Opus 5.5 (breaking changes)
- Anthropic docs: Pricing (cache writes, cache reads, fast mode)
- tbench.ai: Terminal-Bench 4.1 leaderboard