Setup guide · As of 2026-11-22

Cline v4.1.27 & Roo Code Setup for Claude Opus 5.5 and Fable 5.2: Dual-Protocol Base URL Workflow

As of November 22, 2026, to call claude-opus-5-5, claude-fable-5-2, or claude-sonnet-5-5 in Cline (latest v4.1.27) or Roo Code, set API Provider to Anthropic, check Use custom base URL, and enter https://api.qcode.cc/api (no /v1 suffix); to call gpt-6.2-sol, deepseek-v4.1-flash, or glm-5.3, select OpenAI Compatible with https://api.qcode.cc/openai/v1 so one API key powers a dual-model plan-and-execute workflow.

Updated 2026-11-22

Point this tool at the QCode endpoint — one key for Claude, GPT and Chinese models.

#Cline v4.1.27#Claude Opus 5.5#Fable 5.2#Dual Base URL

Four values to check first

v4.1.27

Shipped 2026-11-21: auto-refreshing model lists

Cline v4.1.27 shipped on 2026-11-21: saving an API key or base URL refreshes the provider's model list and shows rejected keys directly in the model picker; read_files is no longer cut to 8k characters and passes up to 2000 lines or about 48k characters.

/api

No /v1 for Anthropic; use /openai/v1 for OpenAI Compatible

All Claude 5 models require the Anthropic provider with Use custom base URL checked and https://api.qcode.cc/api (no /v1); GPT, DeepSeek, and GLM use OpenAI Compatible with https://api.qcode.cc/openai/v1 (no trailing slash).

$4 / $20

Opus 5.5 official rates; Fable 5.2 is $8 / $40

Official per-1M-token API list prices: Anthropic claude-fable-5-2 at $8 input / $40 output, claude-opus-5-5 at $4 input / $20 output, claude-sonnet-5-5 at $2 input / $10 output; OpenAI gpt-6.2-sol at $2.00 / $10.00; Z.AI glm-5.3 at $1.4 / $4.4.

2026-05-15

Roo Code repo is archived; enter new model IDs manually

The GitHub repository RooCodeInc/Roo-Code is marked archived: true with its last push on 2026-05-15, leaving built-in dropdowns on older lists; Roo Code users must enter custom model IDs manually or move to actively updated Cline.

Why custom setups hit 400/401/404: protocol paths and archived presets

As of November 22, 2026, VS Code developers wiring claude-opus-5-5 and claude-fable-5-2 into Cline or Roo Code usually hit three pitfalls. First, protocol mismatch: placing a Claude model ID under OpenAI Compatible (/openai/v1) or appending /v1 to the Anthropic custom base URL causes rejected requests or path errors.

Second, an unchecked box: Cline's Anthropic provider targets the official endpoint by default until you explicitly check "Use custom base URL" and enter the custom URL. Third, stale presets: RooCodeInc/Roo-Code has been archived since 2026-05-15, so its built-in list lacks Claude 5.5, Fable 5.2, and GPT-6.2 Sol, triggering "Model Not Found" unless you type the exact model ID manually.

Splitting Anthropic and OpenAI Compatible into two profiles lets you switch between flagship reasoning and lightweight execution models cleanly.

2026-11-21: three key fixes in Cline v4.1.27

Cline shipped v4.1.27 on 2026-11-21 with three changes that directly improve custom base URL setups and large-repo coding.

  1. 1Auto-refresh and picker errors: the release notes state verbatim, "A provider's model list refreshes after you save its API key or base URL, and a failed refresh (such as a rejected key) shows as an error in the model picker." Invalid keys now surface immediately instead of stalling silently.
  2. 2Full read_files output: read_files results are no longer truncated at 8k characters and now pass in full up to the tool's limit of 2000 lines or about 48k characters, cutting round trips on large files.
  3. 3Reasoning effort and custom overrides: the reasoning effort dropdown now displays the effort a new task will actually use with a Default entry to clear saved settings, and setting reasoning to None now reaches custom models with saved Model Configuration overrides so tool calls are no longer rejected.

Three key dates: Roo Code archived, Cline v4.1.27, and dual-protocol verification

2026-05-15

The official GitHub repository RooCodeInc/Roo-Code received its last push on 2026-05-15 and is marked archived: true. Its built-in model presets predate claude-opus-5-5, claude-fable-5-2, and gpt-6.2-sol, so developers staying on Roo Code must type exact model IDs manually.

2026-11-21

Cline shipped v4.1.27: saving an API key or base URL auto-refreshes the model list and surfaces errors in the picker, read_files no longer cuts at 8k characters (passing up to 2000 lines or about 48k characters), reasoning overrides and custom model tool calls were fixed, and undici was upgraded to 7.30.1.

2026-11-22

Dual-protocol endpoints and vendor pricing verified: QCode's live lineup covers claude-fable-5-2, claude-opus-5-5, and claude-sonnet-5-5 alongside gpt-6.2-sol, deepseek-v4.1-flash, glm-5.3, and kimi-k3.1-pro, all accessible with one cr_ key across Anthropic and OpenAI Compatible modes.

What is confirmed vs common misconceptions and unpublished details

Confirmed (Cline docs, v4.1.27 release notes, GitHub repo metadata, and vendor pricing pages)

As of November 22, 2026, each item below is verified against primary sources.

  1. 1Anthropic provider UI: in Cline Settings, pick Anthropic under API Provider, paste your key into Anthropic API Key, check Use custom base URL, and enter your custom endpoint; after picking a Claude model, Enable Extended Thinking appears below the dropdown, and the docs note you are billed for the full thinking tokens, not just the summary.
  2. 2OpenAI Compatible UI: pick OpenAI Compatible under API Provider, enter Base URL, API Key, and Model ID, and optionally tune Max Output Tokens, Context Window, Image Support, tool calling, and input/output prices in Model Configuration; official troubleshooting lists "Invalid API Key" and "Model Not Found".
  3. 3Cline v4.1.27 (2026-11-21): saving a key or base URL refreshes the model list, read_files passes up to 2000 lines or about 48k characters, compaction no longer falls back to the basic summary on uncataloged custom models, and reasoning text from a cancelled or failed turn no longer leaks into the next turn.
  4. 4Roo Code status: RooCodeInc/Roo-Code is archived: true with last push 2026-05-15.
  5. 5Official vendor API prices per 1M tokens: Anthropic lists Claude Fable 5.2 at $8 / $40 ($0.20 cache read), Claude Opus 5.5 at $4 / $20 ($0.20 cache read), and Claude Sonnet 5.5 at $2 / $10 ($0.10 cache read); OpenAI Standard lists gpt-6.2-sol at $2.00 / $10.00 ($0.10 cached input); Z.AI lists GLM-5.3 at $1.4 / $4.4 ($0.26 cached input); DeepSeek lists DeepSeek-V4.1-Flash-1017 at $0.3 input / $1.2 output peak and $0.15 / $0.6 off-peak (cache hit $0.006 peak / $0.003 off-peak).

Common misconceptions and unpublished vendor details

Avoid these five misconceptions that cause repeated errors or billing surprises.

  1. 1"Put Claude models into OpenAI Compatible so one URL handles everything" — false; all Claude 5 models require the Anthropic endpoint https://api.qcode.cc/api (no /v1), and sending them to /openai/v1 triggers errors or Model Not Found.
  2. 2"Anthropic custom base URLs should end with /v1" — false; Cline appends the request path itself, so adding /v1 after https://api.qcode.cc/api duplicates the path and causes 404 or 400 errors.
  3. 3"Roo Code will soon ship an update adding Opus 5.5 and Fable 5.2" — RooCodeInc/Roo-Code is archived: true with no announced future release, so do not wait for built-in dropdown updates.
  4. 4"Claude Code subscription mode in Cline behaves identically to API mode" — Cline's official docs note that Claude Code subscription mode uses subscription limits instead of token billing, responses may not stream token-by-token, and both image uploads and prompt caching are limited.
  5. 5QCode's live per-model rates are not derived from vendor list prices above; always check qcode.cc/models for live rates.

Single flagship model for every step vs dual-protocol plan-and-execute split

A: Single flagship model end to end (Fable 5.2 or Opus 5.5)

Fastest to configure with one Anthropic profile covering repo reading, edits, and tests; however, mechanical batch edits, log triage, and boilerplate consume flagship tokens at full rate (Anthropic official list prices: Fable 5.2 at $8 input / $40 output, Opus 5.5 at $4 / $20), burning quota quickly in active repos.

B: Dual-protocol split (Claude 5.5/5.2 for planning, Sol / DeepSeek / GLM for execution)

Use Anthropic mode with claude-fable-5-2 or claude-opus-5-5 for architecture design, cross-file refactoring, and code review; switch to claude-sonnet-5-5 (Anthropic official $2 / $10) or OpenAI Compatible with gpt-6.2-sol (OpenAI official $2.00 / $10.00), glm-5.3 (Z.AI official $1.4 / $4.4), or deepseek-v4.1-flash (DeepSeek official peak $0.3 / $1.2) for bulk edits and unit tests. One API key lets you switch in seconds, balancing deep reasoning with lower routine token spend.

Five steps: wire Cline v4.1.27 and Roo Code for both protocols

  1. 1Check the extension version: update Cline in VS Code to v4.1.27 (released 2026-11-21) to get auto-refreshing model lists and full 48k-character read_files output; Roo Code users can skip straight to steps
  2. 2and ④ and enter model IDs manually. ② Configure Anthropic mode (for Claude 5): click the settings icon in the Cline panel, choose Anthropic as API Provider, paste your cr_ key into Anthropic API Key, check Use custom base URL, and enter https://api.qcode.cc/api (use https://asia.qcode.cc/api in mainland China or https://us.qcode.cc/api in North America and Europe; do not append /v1).
  3. 3Pick your Claude flagship and thinking option: select claude-opus-5-5, claude-fable-5-2, or claude-sonnet-5-5 under Model, optionally check Enable Extended Thinking below the dropdown for hard architecture work, and save to verify the picker shows no error.
  4. 4Add an OpenAI Compatible profile (for execution models): create a second profile or switch API Provider to OpenAI Compatible, set Base URL to https://api.qcode.cc/openai/v1 (with /v1 and no trailing slash; mainland China https://asia.qcode.cc/openai/v1, North America/Europe https://us.qcode.cc/openai/v1), paste the same cr_ key into API Key, and enter gpt-6.2-sol, deepseek-v4.1-flash, glm-5.3, or kimi-k3.1-pro as Model ID; on Roo Code, also set max output and context size in Model Configuration if needed.
  5. 5Run a two-stage verification task: ask claude-opus-5-5 or claude-fable-5-2 in Anthropic mode to inspect key files and draft a refactor plan, then switch to gpt-6.2-sol or glm-5.3 in OpenAI Compatible mode to apply batch edits and tests; if Invalid API Key or Model Not Found appears, double-check whether /v1 was added or omitted on the two endpoints.

One QCode key across both protocols: no monthly lock-in and global payment options

QCode is an independent API access service with no affiliation to Anthropic or OpenAI. One cr_ API key drives both protocols in Cline and Roo Code: set https://api.qcode.cc/api under Anthropic mode (https://asia.qcode.cc/api in mainland China, https://us.qcode.cc/api in North America and Europe) for claude-fable-5-2, claude-opus-5-5, claude-sonnet-5-5, and claude-haiku-5-5, and set https://api.qcode.cc/openai/v1 under OpenAI Compatible mode (https://asia.qcode.cc/openai/v1 in mainland China, https://us.qcode.cc/openai/v1 in North America and Europe) for gpt-6.2-sol, gpt-6.2-astra, deepseek-v4.1-flash, deepseek-v4-pro, glm-5.3, glm-5.3-flash, kimi-k3.1-pro, and qwen3.8-max — no juggling separate vendor accounts and no 5-hour rolling subscription lockouts.

Billing is per token, with live per-model rates on /models. Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days; if you already have a Personal Plan, add a booster pack when you need more quota. Supported payment methods include international cards (Visa / Mastercard), Apple Pay, Google Pay, Alipay, WeChat Pay, Russian local bank cards and SBP (plus CIS bank cards), and multi-network USDT cryptocurrency across 18 networks with a 0% platform fee, usually credited within minutes and up to 12 hours.

Frequently asked questions

Why do claude-opus-5-5 or claude-fable-5-2 return 400, 401, or Model Not Found in Cline?

Usually because a Claude model was placed under OpenAI Compatible, /v1 was appended to the Anthropic URL, or Use custom base URL was left unchecked. All Claude 5 models require API Provider set to Anthropic, Use custom base URL checked, and https://api.qcode.cc/api (no /v1). In Cline v4.1.27, saving an API key or base URL refreshes the model list automatically, so rejected keys (HTTP 401) or depleted Pay-as-you-go Pack quota (HTTP 402) surface directly in the picker and logs.

What if Opus 5.5 and Fable 5.2 are missing from the Roo Code dropdown?

Type the exact model ID manually inside a custom provider, or switch to actively maintained Cline v4.1.27. The RooCodeInc/Roo-Code repository is marked archived: true (last pushed 2026-05-15), so its built-in list stays on older presets; once the matching protocol Base URL is set, entering claude-opus-5-5, claude-fable-5-2, gpt-6.2-sol, deepseek-v4.1-flash, or glm-5.3 manually works directly.

Does a Pay-as-you-go Pack auto-renew monthly, and can I keep the same key after 7 days?

No monthly auto-renewal: a Pay-as-you-go Pack is a one-time quota valid for 7 days with a concurrency limit of 5, and buying another pack after expiry restores and adds quota on the same API key (exhausted quota returns HTTP 402, expired packs return HTTP 401, and packs are non-refundable after purchase). Personal Plan keys behave differently: they run on a monthly cycle and cannot be renewed on the original key after expiry, requiring a new plan key.

Do I need to edit code or reinstall Cline/Roo Code to switch an existing setup to QCode?

No code edits or extension reinstalls are needed — just update two settings fields. For Claude, keep the Anthropic provider, check Use custom base URL with https://api.qcode.cc/api, and paste your cr_ key; for GPT-6.2 Sol, DeepSeek V4.1 Flash, or GLM-5.3, point OpenAI Compatible Base URL to https://api.qcode.cc/openai/v1 with the same cr_ key. Existing workspace sessions and MCP tool settings stay intact.

How are tokens and caching counted when Enable Extended Thinking is on or read_files loads big files?

Cline's official docs state that Enable Extended Thinking outputs thinking blocks and you are billed for the full thinking tokens, not just the summary, while Claude prompt caching cuts repeated context costs — on Anthropic's official pricing table, cache reads for both claude-opus-5-5 and claude-fable-5-2 list at $0.20 per 1M tokens. Meanwhile, Cline v4.1.27 lifted the 8k-character cut on read_files (up to 2000 lines or about 48k characters), reducing repeat file reads; QCode bills per token at live rates on /models.

How can I top up without an international credit card? Which payment methods work?

QCode accepts Alipay, WeChat Pay, international cards (Visa / Mastercard), Apple Pay, Google Pay, Russian local bank cards and SBP (plus CIS bank cards), and multi-network USDT cryptocurrency across 18 networks with a 0% platform fee and manual on-chain verification, usually credited within minutes and up to 12 hours. You can buy a Pay-as-you-go Pack or Personal Plan without an overseas credit card.

Sources

Cline v4.1.27 release notes (github.com/cline/cline/releases/tag/v4.1.27, 2026-11-21); Cline official Anthropic and OpenAI Compatible provider docs (github.com/cline/cline/blob/main/docs/provider-config, checked 2026-11-22); Roo Code GitHub repository metadata (github.com/RooCodeInc/Roo-Code, archived: true, last pushed 2026-05-15); Anthropic API pricing (platform.claude.com/docs/en/about-claude/pricing, checked 2026-11-22); OpenAI API pricing (developers.openai.com/api/docs/pricing, checked 2026-11-22); DeepSeek API pricing (api-docs.deepseek.com/quick_start/pricing, checked 2026-11-22); Z.AI pricing (docs.z.ai/guides/overview/pricing, checked 2026-11-22).

One cr_ key for Claude 5.5/5.2 and execution models

No monthly plan required: create a QCode cr_ key and paste it into your existing Cline or Roo Code Anthropic (https://api.qcode.cc/api) and OpenAI Compatible (https://api.qcode.cc/openai/v1) profiles, start with a 7-day Pay-as-you-go Pack, and check live per-model rates on /models.

Not sure where to start?

Try it for a week

Small Pack: $4 for $30 of quota, valid 7 days

Use it every day

Starter: $9.00/month with $5 of quota every day — just $0.060 per $1 of quota, vs about $0.113 for the Medium Pack