Claude Opus 5 Complete Guide
Anthropic's model page puts Claude Opus 5 under legacy models: it is still available, but you should consider migrating to Claude Opus 5.5. Same price and the same context window as Opus 4.8, so you can switch freely; what differs is prompt caching and the knowledge cutoff — the threshold drops from 1,024 tokens to 512, which brings short system prompts and tool definitions into cache range.
Claude Opus 5 (claude-opus-5) has been available since 2026-07-24 at $5 input / $25 output per 1M tokens — the same list price as Opus 4.8. The official model page now lists it under legacy models: still usable, but you should consider migrating to Claude Opus 5.5, and the deprecations table gives a tentative retirement date of no sooner than 2027-07-24. As of 2026-10-20 the newest Opus from Anthropic is still Claude Opus 5.5 (released 2026-09-22, $4 / $20, cache reads $0.20). Prefer official docs for capability and pricing details.
The numbers
$5 input and $25 output per million tokens — the same official price as Opus 4.8.
The official model page lists a 1M-token input context for claude-opus-5, with no extra configuration; the [1m] suffix is how you pick a 1M variant of earlier models in Claude Code.
Cacheable from 512 tokens; Opus 4.8 needs 1024
Up to 128K tokens in a single response
What actually differs from Opus 4.8
The short answer: on the official model pages the price, context window and capability lines of the two models match; what differs is the minimum prompt-cache length and the knowledge cutoff.
Price and window: identical
On the official price list the Claude Opus 5 and Claude Opus 4.8 rows carry the same rates: $5 in / $25 out per million tokens, $6.25 cache write and $0.50 cache read, 1M input and 128K output — line for line the same on both models.
Prompt cache threshold: 512 vs 1024
Prompt caching sets a minimum cacheable length per model: the official caching docs put Claude Opus 5 in the 512-token tier, Claude Opus 4.8 in the 1,024 tier and Claude Opus 4.7 in the 2,048 tier (checked 2026-10-20). Below that minimum, content marked with cache_control is not cached at all. Prompt caching has a minimum length, below which content is simply not cached: Opus 4.8's floor is 1,024 tokens and Opus 4.7's is 2,048 — first checked against the official docs on 2026-09-21, tiers unchanged.
Cache rates are the same for both models on the official price list — $6.25 writes and $0.50 reads per million tokens — so only the length line changes. Agentic workflows with a short system prompt, many tool definitions and dense turns feel that difference most.
Capabilities: the same set
The capability lines match on both official model pages: text and image input, text output, 1M context and 128K maximum output; vision input, function calling, PDF input, computer use and prompt caching are flagged identically on both models, so integration code written for Opus 4.8 usually needs no changes. What differs is the reliable knowledge cutoff — May 2026 for Opus 5 against January 2026 for Opus 4.8.
How to call it
Switch inside Claude Code
Use the /model command mid-session; no restart required.
/model claude-opus-5Set it as the default via environment variable
Useful when you always want one model — put it in your shell profile to make it stick.
ANTHROPIC_MODEL=claude-opus-5 claudeThe 1M context window and the [1m] suffix
The context window on the official model page is already 1M tokens, and Claude Code docs state that Opus 4.7 and later run with the 1M window by default, so no [1m] suffix is needed; the suffix is how you pick a 1M variant for earlier models.
/model claude-opus-5[1m]Choosing across the Claude lineup
The input and output prices per million tokens below are taken line by line from Anthropic's official pricing page; QCode bills per token, with per-model rates listed on /models.
| Model | Input $/M | Output $/M | Context / output | Best for |
|---|---|---|---|---|
| claude-opus-5 | $5.00 | $25.00 | 1M / 128K | Listed by Anthropic as legacy; its lower cache minimum suits tool-heavy, turn-dense work; succeeded by Opus 5.5 on 2026-09-22 |
| claude-opus-4-8 | $5.00 | $25.00 | 1M / 128K | Same price and window; still a row on the official price list |
| claude-fable-5 | $10.00 | $50.00 | 1M / 128K | Priced above Opus, aimed at long-form and creative writing |
| claude-sonnet-5 | $2.00 | $10.00 | 1M / 128K | The value pick for everyday coding |
| claude-haiku-4-5 | $1.00 | $5.00 | 200K / 64K | Formatting, simple Q&A, bulk operations |
Prices in the table were checked line by line against Anthropic's official pricing page. The full QCode list is on the models and pricing page
Frequently asked questions
Should I use Opus 5 or Opus 4.8?
They cost the same and share the same window, so the two are interchangeable, and Opus 4.8 is still marked Active (legacy) and remains in service. The difference is the minimum cacheable prompt length: Opus 5 sits lower, so tool-heavy workflows with short turns are likelier to get cache hits. If you are still choosing, the official model overview says to start with Claude Opus 5.5, the newest Opus since 2026-09-22 ($4 / $20); as of 2026-10-20 that has not changed.
Do I need code changes to migrate from Opus 4.8?
No. Both models carry the same capability and price lines on the official model pages, so usually you only change the model name from claude-opus-4-8 to claude-opus-5.
Which model names are available?
The official model page gives the Claude API name as claude-opus-5, and QCode lists this ID as callable too. In Claude Code you can also append the [1m] suffix to a model name or alias to pick a 1M-token variant.
What does it cost, including caching?
$5 input and $25 output per million tokens; cache writes cost $6.25 and reads $0.50 per million, while the one-hour cache write tier is $10. QCode bills per token — /models carries the per-model rates and the live figures.
What capabilities does it support?
The official model page lists text and image input with text output, and prompt caching is available for this model: vision input, function calling, PDF input, computer use and prompt caching are the same set as Opus 4.8. On the Message Batches API, with the official beta header, a single response can reach 300k output tokens.
Does it need to be enabled separately?
No. Every listed model is callable on every plan; model access does not differ by plan, and there is no unlock fee or application. Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days.
Using Opus 5 on QCode
One key across clients
The same API key works in Claude Code and the OpenAI Codex CLI — no separate purchases.
Every model open
All plans can call all models. Access is not tiered by plan and there is no unlock fee.
Billed per token
QCode bills per token; per-model rates are listed on the /models page.
Direct access from mainland China
Usable directly on mainland networks, with no network workarounds needed on your side.
Try Claude Opus 5 now
One key across Claude, Codex and Chinese models. Personal Plans from $9/mo give a daily quota valid for 30 days, and once activated you can start calling; if a day runs short, buy a booster pack. With no personal plan and no wish for a monthly one, Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days.
Related reading
Claude Opus 5.5 complete guide
The current Opus model from Anthropic; the official guidance is to start there if you are unsure.
Claude Fable 5 complete guide
The tier above Opus, aimed at long-form and creative writing.
Sonnet 5 or Opus 4.8?
Everyday coding on Sonnet, or step up to Opus? Broken down by scenario and cost.