中文
AI Tools / ai-coding

Qoder

Reviewed by AIX Cove ·
Qoder
ai-coding

Qoder at a glance

Qoder is an agentic coding platform built by Alibaba's Qwen team. It launched in August 2025 as a coding assistant, and by 2026 it has grown into a product family that includes an IDE, a JetBrains plugin, a CLI, managed cloud agents, and two AI work apps. The core idea is the same across all of them. Instead of stopping at a suggestion or a chat reply, the agent plans the work, uses tools to execute it, verifies the result, and iterates until the deliverable exists.

I checked the official docs, the pricing page, and the credits guide before writing this. Qoder is free to start and sells credits-based plans. The free tier is genuinely usable, but the features that make it interesting, Quest and Experts Mode, sit behind paid plans. If you just want autocomplete, lighter tools do the same job for less friction.

What the product family actually includes

  • Qoder IDE. A full editor with two workspaces. Editor keeps NEXT code suggestions, Inline Chat, and the Chat panel next to your code for in-flow work. Quest is a separate window for autonomous delegation, with task boards, progress tracking, and artifact review for long-running jobs.
  • JetBrains plugin. Brings code suggestions, Ask, Agent, MCP, and project rules into JetBrains IDEs.
  • Qoder CLI. A terminal-native agent that reads files, runs commands, and commits changes, plus a scriptable engine for automation.
  • Cloud Agents. Managed agents you drive through an API, with configurable environments and streamed sessions.
  • QoderWork and QoderWake. Document, spreadsheet, research, browser, and desktop task delegation, plus always-on agents for recurring work.
  • Enterprise. Centralized purchasing, identity, policies, knowledge, model channels, and audit controls.

On top of these, the IDE adds context features that matter in practice. Repo Wiki generates structured documentation from a codebase and tracks changes to it. Knowledge Cards and Memory accumulate project context across sessions. Rules encode team conventions. MCP lets you connect external tools and services. Experts Mode breaks a large task into parallel sub-agents for planning, implementation, testing, and review. Browser Agent, Computer Use Agent, and Canvas cover web research, GUI automation, and visual annotation.

Pricing and the credits system

Qoder uses credits as its metering unit. Individual plans are Free, Pro at 20 USD per month, Pro+ at 60 USD, and Ultra at 200 USD. New users get a 14-day Pro trial with 300 credits on first sign-in. The Free plan includes limited completions and next edits, a limited daily allowance of basic model calls, and bring-your-own-key support. Paid plans add 2,000, 6,000, and 20,000 credits per month respectively for premium models. When credits run out, Qoder falls back to basic models with a daily limit instead of cutting you off.

Credits burn at very different rates. The official guide gives estimates per request at a 200K context window. Ask Mode in the Editor costs about 4 credits, Editor Agent Mode about 12, Quest Agent Mode about 50, and Quest Experts Mode about 75. Repo Wiki runs about 50 credits per repository. The Pro plan's 2,000 credits disappear fast if you lean on Quest and Experts Mode, which is why the docs recommend Pro+ for heavy agent users and Ultra for people who run long Quest tasks. Credit Packs cost 20 USD for 1,500 credits, are non-refundable, and expire one month after purchase. Plan credits reset to zero at the end of each billing period, so there is no rolling over unused quota.

Typical workflow

Setup is a download and a sign-in. The IDE works on Windows, macOS, and Linux. A typical first day looks like this. Install the IDE, sign in with email, Google, or GitHub, and claim the free Pro trial. Open an existing project, let the agent index it, and ask an Ask Mode question about the codebase to check its understanding. Move to a small real change with Agent Mode and review the diff before accepting it. If the project lacks documentation, run Repo Wiki once to generate a map of modules and entry points. From there, decide whether your usage pattern justifies a paid plan.

For longer tasks, Quest is the more honest test. Hand it a bounded goal, let it work through files and terminal commands in its own window, and review the artifacts it produces. Teams that run these tasks regularly should write Rules for conventions the agent must follow, because a short project instruction beats a growing chat prompt.

Where it fits, and where it does not

Qoder earns its place for developers who want an agentic loop inside an editor rather than a separate CLI tool, and for teams inside the Alibaba or Qwen ecosystem who want tight integration with those models. The IDE is a reasonable Cursor alternative, the JetBrains plugin covers the IDE gap, and the CLI competes with tools like OpenCode and Qwen CLI. Enterprise features make it viable for orgs that need governance on top of agent use.

It is a weaker fit if you only want fast autocomplete, since the free tier is deliberately limited and paid plans are priced for agent workloads. BYOK on the Free plan means you supply your own model keys and pay the provider directly. Teams already standardized on a different agent runtime should weigh the migration cost. And if your need is a local, offline coding agent, Qoder's hosted model access is not that.

Practical tips

  • Spend the trial on Quest, not autocomplete. The 14-day Pro trial with 300 credits is the cheapest way to test Experts Mode and long-running delegation, which are the features that separate Qoder from a plain copilot. Autocomplete alone will not tell you whether the platform fits.
  • Watch the credits burn rate per feature. Quest Agent Mode runs about 50 credits per request and Experts Mode about 75. If you plan heavy autonomous work, budget Pro+ or Ultra, or the monthly quota will vanish in the first week.
  • Generate Repo Wiki on legacy code first. A structured map of an unfamiliar codebase makes every later agent task cheaper and more accurate. It costs about 50 credits per repository and pays for itself quickly.
  • Write Rules before you delegate. Conventions for testing, commit style, and file layout get applied consistently once they are in Rules, instead of being re-explained in every prompt.
  • Connect MCP servers for tools you already use. Qoder speaks the Model Context Protocol, so your existing MCP tooling carries over instead of needing a separate integration.
  • Check the usage page before the end of the billing period. Plan credits reset to zero each month and Credit Packs expire after one month. If you have a balance, spend it on the work you actually need done.

Limits and risks

  • Credits expire on a schedule. Monthly plan credits reset, add-on packs expire after a month, and packs are non-refundable. Light users can lose money to expiry if they buy packs speculatively.
  • The free tier is a teaser, not a daily driver. Limited completions, a daily cap on basic model calls, and no Quest or Repo Wiki mean heavy use requires a subscription.
  • Feature claims come mostly from the vendor. Published customer cases report large gains, like AMAP lifting a one-shot rate from 37.3% to 61.5% and XPeng reporting 96x risk-screening efficiency. Those are vendor-published engineering stories, not independent benchmarks.
  • Independent discussion is still thin. Hacker News coverage of Qoder's launch and Quest Mode was modest compared to Cursor or Claude Code. Most community material is in Chinese and much of it is vendor-adjacent, so cross-check claims against your own workload.
  • Trial restrictions. The 14-day Pro trial requires the latest client and does not work on virtual machines, and the offer is limited to one account per user.

Related tools on AIXCove

Sources and further reading

Sources: official docs & pricing pages, hands-on testing where noted, and community feedback. Prices verified August 2026 and may change.