Qwen Provider (Alibaba DashScope)

What it is

Omniscio supports spawning Claude-Code-style sessions backed by Qwen's Anthropic-compatible API — Alibaba's DashScope endpoint (hosted at dashscope-intl.aliyuncs.com/apps/anthropic) — as an additional provider, alongside claude (the default), codex, gemini, antigravity, deepseek, kimi, glm, minimax, and meta. It is the sixth member of the anthropic-compat family.

A Qwen session looks identical in the sidebar and main pane to a Claude session — same status dots, same streaming bubbles, same Ctrl+Enter to send, same Plan / Auto-Accept / Bypass Permissions modes, same tool-call approval UI. The provider is invisible in normal chat. The only user-visible difference is the Qwen icon (a violet monogram mark, no text label) in the session header.

Tool-call behavior is identical to Claude because the same claude CLI binary is doing the talking — only the model and the upstream endpoint differ. There is no yolo / auto-approve mode like Gemini or Anti-Gravity; the standard Claude approval flow applies.

Where to find it

Everything you configure for Qwen lives in Settings → Accounts — but only once the Show alternative AI providers master toggle is on, which reveals the Qwen panel alongside its siblings. The provider chooser for a single launch sits in the launch-config pickers on a fresh (zero-message) session; the standing per-project default is set from the project's three-dot menu → Edit → Default provider. Once a Qwen session is running, the only place the provider shows is the Qwen icon in that session's header.

How it behaves

How to enable

Master toggle required first. Qwen is an alternative provider — by default Omniscio ships as a Claude-only product, and the entire Qwen setup section below is hidden in Settings → Accounts. Flip Settings → Accounts → Show alternative AI providers to ON and the Qwen panel appears alongside Gemini, Codex, Anti-Gravity, DeepSeek, Kimi, GLM, MiniMax, and Meta. With the master off, the per-project "Default provider" radio collapses to a single Claude row and the session header's provider switcher (ChangeProviderButton) hides Qwen — Qwen effectively does not exist in the UI even if every gate below is configured. Your saved API key and "Allow Qwen sessions" toggle are preserved across master-toggle flips.

You need the toggle below plus a way to pay for Qwen — your own DashScope API key, Omniscio credits, or both, with Qwen's supply list (Settings → Accounts → Who pays & who serves) deciding which pays first (note: no binary check — same claude binary as Claude). The opt-in toggle leads and gates the key field — the API key input appears only once Qwen is turned on, so a key can't be entered into a disabled provider (when off, a hint keeps the qwen-api-key Settings-search anchor landable and a saved key is preserved):

1. Settings → Account → Allow Qwen sessions in any project — opt-in toggle, off by default; same security stance as the Codex/Gemini/DeepSeek/Kimi/GLM/MiniMax/Meta spawn guards. Turning it on reveals the key field below.
2. Settings → Account → Qwen → API key — paste a DashScope API key from the Alibaba Cloud Model Studio console. Encrypted at rest via Electron safeStorage. Optional when Qwen's supply list has an Omniscio credits row (a new user's list has one) — with no key saved, the session runs on your prepaid credits through the gateway. When both are present, the order of Qwen's list decides which pays first. See model-vendors.md.

Qwen needs a credential — your own key or Omniscio credits. With neither, a Qwen spawn is refused up front with a clear "add a key" message (it never crashes or silently falls back to Claude). Qwen is not user-key-only: it rides the same company-credits gateway lane as its Kimi / MiniMax / Meta siblings. When the Omniscio credits row of Qwen's supply list is the row that serves, the child is pointed at Omniscio's gateway instead of DashScope (lane /v1/qwen, funded by the pooled OpenRouter key and mapping each Omniscio qwen/<model> lane onto OpenRouter's own openrouter/qwen/<model> slug) and authenticates with the gateway token rather than a DashScope key. Whichever of the two comes first in Qwen's list pays first, and either credential makes the provider spawn-ready. See FUNDABLE_PROVIDERS in fundable-providers.ts.

Once both are green, you can launch a Qwen session in three ways:

• Per-launch override — on a fresh (zero-message) session, the ChangeProviderButton in the new session's main panel (the launch-config pickers on a fresh session) lets you pick a one-off provider before sending the first message. Non-ready providers appear aria-disabled with an "Add key…" / "Enable in Settings…" suffix and deep-link to the right Settings panel.
• Per-project default — Edit Project dialog (three-dot menu → Edit) has a "Default provider" section listing every enabled provider. Pick Qwen and the project's sidebar "+ New Session" button spawns Qwen automatically.
• Programmatically — anything that creates a session with provider: 'qwen' (recipes, agent-driven sessions, the CLI control HTTP API). Same readiness gates apply on the backend.

Choosing the model

A fresh Qwen session lets you pick the model in the launch-config pickers (next to the provider chooser). The list is registry-driven (MODELS_BY_PROVIDER.qwen in provider-models.ts) and offers five, cheapest first:

• Qwen3.8 Flash (qwen3.8-flash, the default) — Alibaba's Aug-2026 cheap-fast Flash MoE (early Qwen4 architecture); the low-cost, high-throughput tier.
• Qwen3.8 Max (qwen3.8-max) — the flagship top-tier model for the hardest agentic reasoning + coding work. This id is a floating alias — Alibaba may point it at whichever checkpoint is current.
• Qwen3.8 Max 0902 (qwen3.8-max-0902) — the pinned 2026-09-02 checkpoint of that flagship (2.4T-parameter MoE, 1M context), shipped as a coding + agentic refresh. Pick this one when you want a reproducible model that will not shift under you; it bills at the same rates as qwen3.8-max.
• Qwen3.7 Plus (qwen3.7-plus) — the mid tier, and Alibaba's own recommended model for coding tools ("balanced performance and cost, with full tool calling and a 1M-token context window for large codebases"; it is also the Coding Plan's default). Note this one is a 3.7 id, deliberately kept for its coding tuning rather than as a newer generation — a middle price point with proven agentic tool use.
• Qwen3.8 27B (qwen3.8-27b) — the newest open-weight dense model (Aug 2026): 1M context, tool calling, and the cheapest genuinely strong coder on the lane. It is the tier that stops the list jumping straight from Flash to Max.

On the newest generation. There is nothing newer than the 3.8 line to offer: Qwen4 has not shipped (no Max, Flash or Coder variant exists, and the open-weights qwen3.8-flash-next is only a preview of its architecture), and the Qwen3-Coder line's newest entry predates the 3.8 generation. All five ids above are Alibaba-hosted on Model Studio and advertise function calling, so any of them can drive an agentic session.

Use default spawns qwen3.8-flash rather than omitting the model flag — Qwen requires a real Qwen model id (the shared claude binary's built-in default is a Claude id that DashScope rejects as "model not found"), so the registry pins one via getProviderDefaultModel. Qwen has no separate "reasoning effort" knob in Omniscio. A Qwen session never inherits your Claude default model (those are Claude ids its endpoint would reject).

qwen3.8-flash-next is intentionally NOT offered. Qwen's open-weights preview variant (qwen3.8-flash-next) is a self-host / download release, not a first-class DashScope-served API id, so it is left out of the picker. Only the five DashScope-served ids above are selectable.

Per-project default

Each project remembers a default provider in the projectDefaultProviders setting (a Record<projectId, ProviderId>). Empty by default — every project falls back to Claude. Change it from the project's three-dot menu → Edit → Default provider radios; the change persists immediately.

When you set a project's default to Qwen but Qwen isn't ready (toggle off or key missing), an amber "!" badge appears on that project's sidebar row. Hover the badge for a tooltip explaining the gap; click it to deep-link straight to the right Settings panel. The badge re-checks on every render, so fixing the gap clears it without a refresh. Claude defaults never show the badge — Claude is always considered ready.

Error states and fixes

Two gap codes — fewer than Gemini/Codex because there is no separate binary to check:

Gap	What it means	Click-to-fix lands you at…
toggle-off	"Allow Qwen sessions" is OFF in Settings	Settings → Account → Allow Qwen sessions toggle
key-missing	No row of Qwen's supply list can serve (no Qwen key, and no other row ready)	Settings → Account → Qwen API key field, or Who pays & who serves

Test your key. The Qwen API-key field has a "Test key" button that pings DashScope's models list (dashscope-intl.aliyuncs.com/compatible-mode/v1/models) with your key, returning one plain-language verdict (works / invalid key / out of balance / model not available) — so a misconfigured key surfaces before you spawn a session rather than as a confusing mid-turn failure.

Retired / unavailable model. If a session's selected model is no longer accepted by DashScope — either Omniscio has removed it from the picker or the vendor killed a still-listed id server-side — Omniscio surfaces a clear "this Qwen model is no longer available — open the model picker and choose another" and parks the session as needs_you / recovery_failed, instead of silently retrying the rejected request into the generic placeholder-stuck give-up loop. This shares the same anthropic-compat guards the DeepSeek/Kimi/GLM/MiniMax/Meta siblings use. See provider-spawn-model-availability-contract.md.

The real reason, not a blanket "model unavailable." A Qwen rejection arrives the same way regardless of cause — a synthetic placeholder with zero output — so when the rejection is not about the model (the account ran out of balance, the conversation outgrew the context window, the key became invalid, or you hit a rate limit), Omniscio reads DashScope's actual error off that placeholder and surfaces the specific cause + fix: out of balance → top up the account, then resend; conversation too long → start a fresh session; bad/expired key → check the Qwen key in Settings; rate-limited → wait a moment, then resend. This shares the exact classifyAnthropicCompatPlaceholderError path the DeepSeek/Kimi/GLM/MiniMax/Meta siblings use.

Cost tracking

Spend hits the standard api_cost_log table because the claude CLI emits per-turn cost events in its stream-json output and the spawn layer already plumbs those into the cost tracker. The source field on the cost row carries the provider ID (qwen), so the Stats → Spend dashboard breaks down Qwen spend separately from Claude/Codex/Gemini.

Cost is re-priced, not taken verbatim. The claude binary computes its total_cost_usd from its OWN (Claude) price table, which has no Qwen row — so it would bill Qwen at Claude rates. Qwen is therefore costReporting: 'estimated', and Omniscio recomputes each turn's cost from the token counts × Qwen's real per-M rate via repriceAnthropicCompatTurnCost — the same token-pricing path all five sibling anthropic-compat vendors use. Rates (model-pricing.ts MODEL_PRICING, all per 1M tokens at Alibaba's Singapore/International list — the region dashscope-intl serves): qwen3.8-flash $0.16 in / $0.47 out; qwen3.8-max and its pinned checkpoint qwen3.8-max-0902 $2.00 in / $6.00 out each, with a $0.25 implicit-cache-read rate (DashScope's published table, confirmed 2026-08-26; the 0902 checkpoint re-confirmed against Model Studio's price page 2026-09-02); qwen3.7-plus $0.40 in / $1.60 out; qwen3.8-27b $0.50 in / $3.00 out (both from the same Model Studio price page, confirmed 2026-09-19). See central-ai-spend-contract.md.

Telemetry: every successful Qwen spawn fires the spawn_non_claude_session feature event, recording { provider: 'qwen' } only — no session ID, project ID, or prompt content.

For agents

How it works under the hood

Spawn-env redirect, no alternate manager. DashScope publishes an Anthropic-compatible endpoint — same request/response shape as Anthropic's Messages API, just at a different base URL. So Omniscio reuses the existing claude CLI binary (the same one Claude sessions use) and redirects every HTTP call at spawn time by setting two environment variables on the child process:

• ANTHROPIC_BASE_URL=https://dashscope-intl.aliyuncs.com/apps/anthropic — sends every HTTP request to DashScope's compat endpoint instead of api.anthropic.com. Like DeepSeek/MiniMax (and unlike Meta's bare host), the base carries an /apps/anthropic path; the claude CLI appends /v1/messages. The base URL is env-overridable via AMC_QWEN_BASE_URL to point at a different DashScope region (US Virginia dashscope-us.aliyuncs.com/apps/anthropic; Beijing dashscope.aliyuncs.com/apps/anthropic).
• ANTHROPIC_AUTH_TOKEN=<your Qwen API key> — supersedes the Claude OAuth/API-key for this child only.

The child process never knows it isn't talking to Anthropic. Streaming, tool calls, --resume for multi-turn, --output-format stream-json parsing all "just work" because they're handled by the standard claude binary against an upstream that speaks the same protocol.

No alternate session manager exists for Qwen — readiness gates, the spawn_non_claude_session feature event fires, then it falls through to the same createSessionWithPrompt() path Claude sessions use. The provider: 'qwen' value is stored on the session row in the database; spawn-cluster-manager.ts reads that column and injects the right spawn-env on every CLI invocation (initial spawn, resume, aside-runner).

Because Qwen runs the claude binary, it is eligible for the same crash/restart/suspended/re-arm recovery as native Claude (claude --resume re-applies the Qwen env overlay from the session row) — distinct from Anthropic account machinery (rate-limit recovery, load-balancing), which it stays out of since it authenticates with your Qwen key, not an Anthropic login.

Data locality (a China-vendor note). The default dashscope-intl endpoint is Singapore-hosted, but Alibaba is a China-based vendor (the same jurisdiction as the OpenRouter → Alibaba helper leg). Prompt content sent from a Qwen session is processed by Alibaba under your own DashScope account. Qwen is listed as a cross-border subprocessor disclosure in SUBPROCESSORS.md (row #69).

Files

• src/shared/providers/registry.ts — the qwen descriptor (anthropic-compat runtimeKind, capabilities, label, pickerOrder 22)
• src/main/services/providers/main-registry.ts — the Qwen readiness wiring (toggle + API-key gates, no binary gate)
• src/shared/providers/provider-setup-catalog.ts — the dashscope-intl.aliyuncs.com/apps/anthropic endpoint, models-list URL, and AMC_QWEN_BASE_URL region override
• src/main/services/anthropic-compat-provider.ts — the base URL + credential accessor (ANTHROPIC_COMPAT_PROVIDERS membership)
• src/main/services/qwen-credential-store.ts — the encrypted qwenApiKey store
• src/main/db/migrations/20260826201344-allow-qwen-provider.ts — extends the sessions.provider allow-list to include qwen
• src/main/process/spawn-cluster-manager.ts — injects ANTHROPIC_BASE_URL + ANTHROPIC_AUTH_TOKEN per session row's provider column
• src/shared/types/provider-readiness.ts — ProviderId union
• src/shared/types/settings/accounts-providers-settings.ts — allowQwenSessionSpawn setting
• src/shared/types/settings/ai-features-settings.ts — qwenApiKey setting
• src/renderer/src/features/settings/sections/accounts/AnthropicCompatProvider.tsx — the shared anthropic-compat settings card; Qwen's per-provider config was added here (shown under "Show alternative AI providers")
• src/renderer/src/components/ui/QwenIcon.tsx — the Qwen monogram icon
• gateway/litellm/config.yaml — the qwen/<model> lane rows a company-credit session routes through; a picker model with no row here stalls with no upstream
• tests/unit/lint/company-credit-lane-routes.test.ts — locks the picker ↔ litellm-route invariant for this lane and every other credits-funded lane

Related

Qwen shares its shape with a whole family of Anthropic-compatible providers, and the pages for its siblings — meta-provider.md, minimax-provider.md, glm-provider.md, kimi-provider.md and deepseek-provider.md — describe the same arrangement with a different vendor and key; Meta is the fifth of them and the one that differs, carrying a bare-host base URL. switching-providers.md is the unified session-provider guide for how a session's provider is chosen and changed, model-vendors.md explains who pays for a Qwen session (its supply list), and the invariants all six of the siblings share are locked in provider-registry-contract.md.