skip to content
back to blogs

Arc: v0.6.2 Release

Arc version 0.6.2, first released on August 28, 2026.

Highlights

Per-Domain Auto Routing

Auto mode now classifies each prompt by domain: code, math, deep reasoning, or agentic work. Classification uses a new domain head and completes in under one millisecond. Per-domain capability thresholds replace the previous single threshold.

Model selection now accounts for reliability, latency, provider health, and cost together, and selects the lowest-cost model that clears the threshold. The difficulty model was retrained on the full 6,290-prompt labeled set. AUC increased from 0.757 to 0.782. When confidence is high, Auto mode sends without confirmation. When confidence is low, it requests confirmation. Learn more.

Checkpoint and Context Rewind

Reverting to a message now restores the conversation and files, including points before a compaction boundary. Checkpoint snapshots taken in the same turn now merge instead of overwriting each other. Restores are fault-tolerant per file and report an error list. Rewinding to a pre-compaction point also restores the archived context.

Tool-Run Summaries

Tool-call chains can now be summarized by configuration. Summaries use either an Arc-generated usage message (for example, "Globbed and searched files") or a short model-written summary.

Changelog

Chat

  • LaTeX math now renders in chat through a self-contained mini-KaTeX.
  • Code blocks now use updated styling within the chat interface.
  • Reopening a chat now restores the stored context percentage and cost. Real provider prompt tokens are persisted instead of re-estimated.
  • The recorded model for each turn now matches the selected model. Costs no longer show as $0 from incorrect attribution.
  • Turns that end with no visible output, including reasoning-only responses, are now retried automatically with a hidden continuation prompt.
  • Tool chains can now be summarized as a count, a usage phrase, or a model-written summary, depending on configuration.
  • Expand no longer clips content in long tool chains.
  • Native notifications now fire only when the VS Code window is unfocused.
  • The top-bar cost now renders to one decimal place. Hovering shows the exact amount to three decimal places.

Sidebar Chat

  • Chat search no longer allows stale results to overwrite fresh results.

Fullscreen Chat

  • None.

Settings

  • Added UI and mono font selection, including custom family inputs.
  • Added a tool-run summary setting with count, top-tools, and model-written summary options.
  • Auto mode quality is now a single preset: balanced, economy, or power.
  • Added arc.router.autoRoute to skip the routed-model confirmation in Auto mode and send directly.
  • The compaction safety margin now applies. The token reserve adapts to the model when no fixed strategy is set.

Backend

  • Auto mode now routes per domain. Selection minimizes a utility function over clearing margin, latency, health, and cost. Routing is limited to models with a real Artificial Analysis score, weighted by live provider health and latency.
  • A per-user tau now raises the routing threshold after soft failures from auto-routed turns.
  • Retrained the difficulty model on the full 6,290-prompt set (AUC 0.757 to 0.782). Rebuilt calibration and capability data from all labels with three reliable anchors. Removed a broken anchor that had reduced routing quality to 0.49. Router assets are now versioned so caches refresh.
  • Requests now send per-provider app-attribution headers, a descriptive user agent, and vendor-prefixed extras.
  • Compaction was reworked. Segmentation no longer miscounts system messages. Re-compaction folds prior summaries into the new summary. noCompact protection covers whole tool chains. Dropped messages are archived and available to the model through context.retrieve.
  • The rewritten condensing prompt now records intent verbatim, decisions, code state, errors, task state, and constraints.
  • Checkpoint snapshots in the same turn now merge instead of overwriting. Restores include files changed through shell or browser tools.
  • Reverting to a message now unions each undone turn oldest-wins per file and searches the archived region. Rewinds restore conversation context and files across compaction. Restores are per-file fault-tolerant with an error list.
  • The 4 MiB stream cap now counts assistant text only. Long-reasoning turns no longer abort mid-stream.
  • Fixed agent lifecycle leaks. Orphaned agents are stopped before replacement. Approval and clarification timers are cleared. Deactivation now kills processes, terminates provider servers, and disposes listeners.
  • Per-token webview broadcasts are now coalesced to about 20 messages per second. Duplicate session posts are removed. Inbound messages now use a size check instead of JSON.stringify.
  • The MCP marketplace cache is now bounded, and its searches run in parallel, alongside browser-tab restore.
  • The chat history codec is now ARCX v2. It decodes v1 archives for backward compatibility.
  • Audit-log lock-contention warnings are now suppressed when transient and rate-limited otherwise.
  • Migrated to pnpm 11.