Claude Code model mix: opusplan and the three aliases
As of 2026-10-13, the official mechanisms that bear directly on how Claude Code divides work across the three tiers are the opusplan alias (opus during plan mode, then it switches to sonnet for execution), the three alias variables ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL and ANTHROPIC_DEFAULT_HAIKU_MODEL, and the model field in a subagent definition; the official docs also list settings such as claude --model and ANTHROPIC_MODEL. The two 5.5 models carry one input and one output price each: Opus 5.5 at $4 / $20 and Sonnet 5.5 at $2 / $10 per million tokens; only Haiku 5.5 has a break at 100K prompt tokens, at $0.10 / $0.50 below it. The qcode.cc guide lists claude-opus-5-5 and claude-sonnet-5-5 as callable models, and QCode has supported claude-haiku-5-5 since 2026-10-09. Billed per token; per-model rates are listed on /models.
Updated 2026-10-13
Point this tool at the QCode endpoint — one key for Claude, GPT and Chinese models.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard / AMEX), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Three official prices and one alias
Opus 5.5 input / output (per million tokens)
Official wording: Opus 5.5 costs $4 per million input tokens and $20 per million output tokens.
Sonnet 5.5 input / output (per million tokens)
Official wording: Sonnet 5.5 is priced at $2 per million input tokens and $10 per million output tokens. On cache hits the official pricing page says both are 5% of the standard input price: $0.20 on Opus 5.5 and $0.10 on Sonnet 5.5 per million tokens; the 2026-10-07 release notes cut Sonnet 5.5 cache reads from $0.20 to $0.10 (0.1x to 0.05x), with cache writes and all other prices unchanged.
Haiku 5.5 (prompts up to 100K tokens)
Official wording: For prompts up to 100K tokens, it is $0.10 per million input tokens and $0.50 per million output tokens. Above 100K it is For prompts over 100K tokens, it is $0.50 per million input tokens and $2.50 per million output tokens, and the docs note a Haiku 5.5 request costs more per token when its prompt is longer than 100K tokens.
One alias that switches on its own
Official wording: Special mode that uses opus during plan mode, then switches to sonnet for execution. Plan mode uses opus for complex reasoning and architecture decisions and execution mode automatically switches to sonnet; opusplan is also selectable in the /model list. To run both phases with 1M context, the official wording sets the model to opusplan[1m], for example with /model opusplan[1m]; setting opusplan[1m] with /model requires Claude Code v2.1.265 or later, and on earlier versions use the --model flag or the model setting instead.
Official mechanisms tied to this three-tier split
As of 2026-10-13 this page covers the two kinds of mechanisms in the official docs that matter directly to this split. First, the opusplan alias: the docs say In plan mode: uses opus for complex reasoning and architecture decisions and In execution mode: automatically switches to sonnet for code generation and implementation, and they add The plan-mode Opus phase uses the same context window as the opus model setting, and the execution phase uses the same window as sonnet. Second, the three alias variables ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL and ANTHROPIC_DEFAULT_HAIKU_MODEL; the docs state Each value must be a full model name, or the equivalent identifier for your API provider, and aliases resolve to different versions by provider, so to pin one version use the full model name (the official example is claude-opus-5-5) or set the matching variable. A subagent can also pin a smaller model with its own model field; running /model with no argument opens the picker.
Release line of the three 5.5 models and one official price cut
Opus 5.5 was released on September 22, 2026; the official release notes say We've launched Claude Sonnet 5.5 (claude-sonnet-5-5), dated September 28, 2026; on October 7, 2026 Claude Haiku 5.5 was released (official ID claude-haiku-5-5), described by the vendor as a model tuned for high-volume and latency-sensitive work. The same day the vendor halved the price of Sonnet 5.5 cache reads, which it says makes Sonnet 5.5 around 20% cheaper on most agentic work (a vendor statement, not a third-party audit). The docs also state that from v2.1.293 the haiku alias resolves to Haiku 5.5 on the Anthropic API.
Releases and price change
Opus 5.5 released, officially priced at $4 input / $20 output per million tokens.
The official release notes say We've launched Claude Sonnet 5.5 (claude-sonnet-5-5); the $2 input / $10 output price per million tokens is on the official Sonnet 5.5 product page.
Haiku 5.5 (claude-haiku-5-5) launches; the same day Sonnet 5.5 cache reads drop from $0.20 to $0.10 per million tokens, with cache writes and all other prices unchanged.
Confirmed vs not verified
Confirmed (verbatim in the official pages)
All of this is verbatim-checkable: the opusplan definition (opus in plan mode and sonnet for execution); the three alias variables ANTHROPIC_DEFAULT_OPUS_MODEL / ANTHROPIC_DEFAULT_SONNET_MODEL / ANTHROPIC_DEFAULT_HAIKU_MODEL with their official descriptions; the subagent model field values sonnet, opus, haiku, fable, a full model ID or inherit; a cache hit being 5% of the standard input price ($0.20 on Opus 5.5 and $0.10 on Sonnet 5.5 per million tokens); the two 5.5 prices $4 / $20 and $2 / $10 plus the Haiku 5.5 break at 100K prompt tokens ($0.10 / $0.50 below it and $0.50 / $2.50 above it); the release dates September 22, 2026 (2026-09-22), September 28, 2026 (2026-09-28) and October 7, 2026 (2026-10-07); Haiku 5.5 with a 1M token context window and 128k max output tokens, plus the official line On the Anthropic API, Haiku 5.5 runs with the 1M context window on every plan, so no [1m] suffix is needed; Sonnet 5.5 requiring v2.1.284 and Opus 5.5 requiring v2.1.280 while v2.1.293 is recommended with Haiku 5.5; and v2.1.265 for setting opusplan[1m] with /model.
Three things the official docs do not say
First, the Claude Code official docs do not say how much work each of the three tiers should carry and do not prescribe a fixed division of labour; the usage advice on this page comes from QCode's own cost-optimization doc, which says to drive the bulk of development with Sonnet, temporarily /model opus to think through an architecture sticking point, then switch back to Sonnet once the decision is made and hand bulk simple cleanup to Haiku. The docs also list settings such as the --model flag and ANTHROPIC_MODEL, while the mechanisms that bear directly on this split are opusplan, the three alias variables and the subagent model field. Second, there is no independently verified capability comparison, and vendor-stated speed and cost figures are labelled as vendor statements. Third, per-model prices on QCode are not listed on this page; the current rates are on /models.
Two routes: opusplan or pinned aliases
opusplan: one alias that switches for you
The official text says The opusplan model alias provides an automated hybrid approach: one alias covering both the planning and the execution phase, opus in plan mode and sonnet for execution, while those two phases still resolve through the opus and sonnet variables (The model to use for opus, or for opusplan when Plan Mode is active and The model to use for sonnet, or for opusplan when Plan Mode is not active). opusplan is selectable in /model; to get 1M context in both phases use opusplan[1m], which needs Claude Code v2.1.265 or later when set with /model, and on earlier versions use the --model flag or the model setting instead.
Pin the three aliases and the subagent model
Finer-grained: set ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL and ANTHROPIC_DEFAULT_HAIKU_MODEL separately, and note the official description says the haiku variable also drives background functionality; a subagent can set its own model (including inherit), and there is a default model for subagents with no model set. These are only mechanical differences; the docs do not say which is better.
How to configure it: six steps
① In a session run /model opusplan to switch at once, or run /model with no argument to open the picker; to run both phases with 1M context, set the model to opusplan[1m], which requires Claude Code v2.1.265 or later when set with /model, and on earlier versions use the --model flag or the model setting instead. ② You can also run /model claude-haiku-5-5 in a session, or start Claude Code from your shell with claude --model claude-haiku-5-5. ③ To pin aliases to specific versions, export ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL and ANTHROPIC_DEFAULT_HAIKU_MODEL with full model names. ④ For subagents, set model in the definition to sonnet, opus, haiku, fable, a full model ID or inherit, or set a default for subagents without a model. ⑤ Check versions: v2.1.293 is recommended with Haiku 5.5, Sonnet 5.5 requires v2.1.284 and Opus 5.5 requires v2.1.280. ⑥ Note Haiku 5.5 costs more per token when the prompt is longer than 100K tokens.
On QCode
Set ANTHROPIC_BASE_URL to https://api.qcode.cc/api (no trailing slash) and ANTHROPIC_AUTH_TOKEN to the API key you create in the console after logging in, which starts with cr_. Then set ANTHROPIC_DEFAULT_OPUS_MODEL to claude-opus-5-5, ANTHROPIC_DEFAULT_SONNET_MODEL to claude-sonnet-5-5 and ANTHROPIC_DEFAULT_HAIKU_MODEL to claude-haiku-5-5. The qcode.cc guide lists claude-opus-5-5 and claude-sonnet-5-5 as callable models; claude-haiku-5-5 has been supported since 2026-10-09 and is called with the same key on the same Anthropic-protocol endpoint https://api.qcode.cc/api by putting the matching ID in model. A subagent model field can also take these full IDs or the haiku alias. Claude models only work on the Anthropic protocol. Top up via Alipay, WeChat Pay, bank cards (Visa / Mastercard / AMEX), Apple Pay / Google Pay (using the card already stored in the wallet) or cryptocurrency (many coins and networks, manual on-chain verification, send a small test amount first); self-service activation takes effect immediately after top-up. Billed per token; per-model rates are listed on /models.
Frequently asked questions
What is opusplan?
It is a model alias named in the official docs: during plan mode it uses opus for complex reasoning and architecture decisions, then automatically switches to sonnet in execution mode. The official line is This pairs Opus’s reasoning for planning with Sonnet’s efficiency for execution. opusplan is selectable in /model; to run both phases with 1M context use opusplan[1m], which requires Claude Code v2.1.265 or later when set with /model, and on earlier versions use the --model flag or the model setting instead.
What do the three ANTHROPIC_DEFAULT variables control?
They decide which model each of the opus, sonnet and haiku aliases resolves to. The official descriptions read: The model to use for opus, or for opusplan when Plan Mode is active; The model to use for sonnet, or for opusplan when Plan Mode is not active; and The model to use for haiku, or background functionality. The docs also state Each value must be a full model name, or the equivalent identifier for your API provider.
How do I make a subagent use a cheaper model?
Set the model field in the subagent definition — the official line is The model field controls which model the subagent uses, and it accepts sonnet, opus, haiku, fable, a full model ID or inherit. The official example is define one with model: haiku to run exploration on a lower-cost model; you can also set the default model for subagents without a model.
What are the official per-token prices of the three models?
The two 5.5 models carry one input and one output price each: Opus 5.5 is $4 input / $20 output per million tokens and Sonnet 5.5 is $2 input / $10 output per million tokens. Only Haiku 5.5 has a break by prompt length: $0.10 input / $0.50 output for prompts up to 100K tokens and $0.50 input / $2.50 output above 100K. The official pricing page says a cache hit on Opus 5.5 and Sonnet 5.5 costs 5% of the standard input price ($0.20 and $0.10 per million tokens), and the 2026-10-07 release notes cut Sonnet 5.5 cache reads from $0.20 to $0.10 per million tokens.
Do I need a specific Claude Code version for Haiku 5.5?
Yes. The docs state that from v2.1.293 the haiku alias resolves to Haiku 5.5 on the Anthropic API, and recommend Use v2.1.293 or later with Haiku 5.5; Sonnet 5.5 requires Claude Code v2.1.284 or later and Opus 5.5 requires v2.1.280 or later.
Which payment methods work for topping up and activating?
You can top up in renminbi with Alipay or WeChat Pay; Visa / Mastercard / AMEX, Apple Pay / Google Pay (using the card already stored in the wallet) and multi-coin, multi-network cryptocurrency also work. Self-service activation takes effect immediately after top-up. Billed per token; per-model rates are listed on /models.
Sources
From Anthropic: the Claude Code official docs pages model-config and sub-agents, the Claude Platform pricing page and release notes, the official Opus / Sonnet / Haiku product pages, the Opus 5.5 and Haiku 5.5 release pages, and claude.com/pricing. From QCode: the docs.qcode.cc pages for the Claude Code tutorial, endpoints and API formats and payment methods, plus qcode.cc /models, the Haiku 5.5 guide page, the Claude Code usage guide page and the cost-optimization doc. All fetched on 2026-10-13.
Three aliases, one key
Point ANTHROPIC_BASE_URL at https://api.qcode.cc/api and set the three ANTHROPIC_DEFAULT variables to claude-opus-5-5, claude-sonnet-5-5 and claude-haiku-5-5, supported on QCode since 2026-10-09. Billed per token; per-model rates are listed on /models.
Related reading
Haiku 5.5 guide
Haiku 5.5's two price tiers, 1M context and the callable ID.
Sonnet 5.5 vs Opus 5.5
A side-by-side of the two models' prices and specs.
Using Claude Code in China
Endpoint setup and Alipay or WeChat Pay top-up.
This page is checked against 2026-10-13. Claude Code aliases, environment variables and the subagent field follow the current Anthropic official docs; prices follow the official pricing page and release notes. QCode endpoints, configuration and payment methods follow docs.qcode.cc. Model availability and per-model rates follow /models and may change without notice. Vendor-stated speed and cost figures were not independently verified and are no promise for any specific workload.