Upgrade Qwen Code to v0.26.3 and treat qwen4 as 1M-context
Qwen Code v0.26.3 has no known breaking changes. Upgrade, then use qwen4 and invite-only qwen-latest-series- aliases as 1M-context models, opt in to memory extraction after no-op runs, and rely on quoted-XML tool-call recovery plus shrinking tool results to auto-compaction headroom.Why this loop
v0.26.3 is an improvement release with no known breaking changes. Context-window detection now treats invite-only qwen-latest-series- aliases as 1M-context models and treats uncatalogued qwen4 variants as 1,000,000-token windows, so sessions still clamped to smaller windows waste capacity. Memory extraction is now an opt-in cadence after a no-op run—the switch to flip if idle turns skip memory. Core recovers outer XML tool calls with quoted call content and shrinks tool results to the headroom left before auto-compaction, so stop hand-truncating tool output and keep XML calls quoted. Persistent session Workflow mode (DW-04A) is the matching change for long-running sessions. Do not assume extra unnamed changes; 31 further notes were omitted from the signal.
Proposed actions
- Install Qwen Code v0.26.3 from https://github.com/QwenLM/qwen-code/releases/tag/v0.26.3, restart the CLI/Web Shell process, and keep existing configs—this release lists no known breaking changes.
- Set this session's model to a qwen4 variant or an invite-only qwen-latest-series- alias. Context-window detection now treats those as 1,000,000-token models when no upstream catalog entry exists, so remove any smaller manual context caps.
- Turn on the opt-in memory extraction cadence that runs after a no-op turn (#13571) so idle/no-op runs still extract memory instead of waiting for a productive tool call.
- Leave large tool outputs untrimmed at the prompt layer. Core now shrinks tool results to the headroom left before auto-compaction (#13599) and recovers outer XML tool calls whose content is quoted (#13579)—keep those calls quoted rather than flattening them.
- Enable persistent session Workflow mode (DW-04A) on this repo's long-running Qwen Code sessions so workflow state is stored with the session instead of being lost on restart.
Agent prompt
Paste into your agent or query via MCP (get_agent_prompt) — free, no extra AI costCopilot Chat / Agent mode
DevAgentRadar → GitHub Copilot
You are helping me adopt a real coding-assistant change. Work only from the facts below. Do not invent features.
Context
Assistant: Qwen Code Proposal: Upgrade Qwen Code to v0.26.3 and treat qwen4 as 1M-context Summary: Qwen Code v0.26.3 has no known breaking changes. Upgrade, then use qwen4 and invite-only qwen-latest-series- aliases as 1M-context models, opt in to memory extraction after no-op runs, and rely on quoted-XML tool-call recovery plus shrinking tool results to auto-compaction headroom. Primary source: https://github.com/QwenLM/qwen-code/releases/tag/v0.26.3
Why it matters
v0.26.3 is an improvement release with no known breaking changes. Context-window detection now treats invite-only qwen-latest-series- aliases as 1M-context models and treats uncatalogued qwen4 variants as 1,000,000-token windows, so sessions still clamped to smaller windows waste capacity. Memory extraction is now an opt-in cadence after a no-op run—the switch to flip if idle turns skip memory. Core recovers outer XML tool calls with quoted call content and shrinks tool results to the headroom left before auto-compaction, so stop hand-truncating tool output and keep XML calls quoted. Persistent session Workflow mode (DW-04A) is the matching change for long-running sessions. Do not assume extra unnamed changes; 31 further notes were omitted from the signal.
Suggested actions
-
Install Qwen Code v0.26.3 from https://github.com/QwenLM/qwen-code/releases/tag/v0.26.3, restart the CLI/Web Shell process, and keep existing configs—this release lists no known breaking changes.
-
Set this session's model to a qwen4 variant or an invite-only qwen-latest-series- alias. Context-window detection now treats those as 1,000,000-token models when no upstream catalog entry exists, so remove any smaller manual context caps.
-
Turn on the opt-in memory extraction cadence that runs after a no-op turn (#13571) so idle/no-op runs still extract memory instead of waiting for a productive tool call.
-
Leave large tool outputs untrimmed at the prompt layer. Core now shrinks tool results to the headroom left before auto-compaction (#13599) and recovers outer XML tool calls whose content is quoted (#13579)—keep those calls quoted rather than flattening them.
-
Enable persistent session Workflow mode (DW-04A) on this repo's long-running Qwen Code sessions so workflow state is stored with the session instead of being lost on restart.
After you finish
Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.
Tell the human: open https://devagentradar.com/proposals/qwen-code-v0-26-3-upgrade-qwen-code-to-v0-26-3-and-treat-qwen4-as-1m-con and mark Applied, Skipped, or Failed. Proposal id: 9c96fec4-9a56-4299-aef8-003dd16c6e26
Your job
- Restate the change in one sentence.
- Propose a minimal plan for my repo (or a throwaway pilot).
- Implement only what I approve; prefer small diffs and tests.
- Call out risks (permissions, breaking APIs, cost).
Start by confirming you understood the proposal.
Your loop
This browser · no sign-in · not shared as “you”After you run the prompt
Only you can mark this. Agents cannot write your loop.
Your decision stays on this device. A public tally appears after a few votes.