Hardcore Reviews
Hardcore Reviews

qwen-code Preview, OpenBot, free-claude-code: Three Compared

Sets the qwen-code v1.0.0-preview beside OpenBot v1.0.2 and the rolling v6.x of free-claude-code in one table, comparing version states and access boundaries rather than picking a winner. Five rows cover models and providers, where AG-UI and MCP attach, permissions and approvals, stability during Preview, and license points. The free-claude-code column states AGPL-3.0-only, the NOASSERTION value from GitHub, the README disclaimer that it is not affiliated with Anthropic, and that provider terms govern integrations. Star counts are cumulative totals.

Published November 15, 20269 min read
<!-- coding-agent-preview-access-boundaries-review | review | qwen-code Preview, OpenBot, free-claude-code: Three Compared -->

Set a Preview build beside two projects that already ship formal tags or rolling version numbers, and the thing worth separating first is not which one has more features, but the fact that three different version states are being read side by side. qwen-code is a Preview whose interfaces may still change, OpenBot has a formal v1.0.2 tag, and free-claude-code keeps a rolling v6.x line with no 1.0.0-style release at all. Three open-source projects are compared here along a single axis: version status and access boundaries. This is not a buying guide, and it hands out no scores and no ranking. A Preview sitting next to two settled releases is the contrast in the title, and it is also an easy thing to read wrong.

As of November 14, 2026, every version number, UTC timestamp, license identifier and feature line quoted below comes from the three repositories' GitHub API responses, README files, LICENSE text and release notes. Star counts are cumulative totals returned by the API, not single-day gains. The angle is worth marking against earlier pieces on this site. Coding agent capability comparison and Terminal coding agent comparison are about capability and selection; Credential and permission comparison and Sandbox isolation comparison focus on permissions and isolation. This article stays on version status and access boundaries and does not repeat a selection verdict; see the in-site articles for those other angles.

1. Three version states should not be read as one

The qwen-code side has two lines to track. The main line is still v0.27.3, tagged at 2026-11-13 00:24:34Z. A separate Preview line starts with the v1.0.0-preview release at 2026-11-11 03:55:44Z, followed by v1.0.0-preview.1 at 2026-11-12 14:25:49Z. The repository is TypeScript, licensed Apache-2.0, created 2025-06-26, and shows 30121 stars and 3417 forks as cumulative totals at the snapshot date. The line that matters is in the release notes, verbatim: Qwen Code 1.0 Preview is now available. Some interfaces, configuration options, and behaviors may change during Preview. In other words v1.0.0-preview is a Preview, not a formal 1.0.0 release; during Preview the interfaces, configuration options and behaviors may change, so it cannot be treated as a stable production build. Reading it as qwen-code having shipped 1.0.0 is an easy misreading this table is meant to prevent.

OpenBot is a different state. The CopilotKit/OpenBot v1.0.2 tag was published at 2026-11-11 08:45:38Z, and the repository carries fifteen tags in total, including v1.0.0 and v0.3.4. It is MIT licensed, created 2026-08-17, with 12126 stars and 1489 forks as cumulative totals. This is a clear formal-tag line: version numbers advance along v1.0.x, and the desktop side ships a notarized macOS dmg plus a signed Windows installer. Set against the qwen-code Preview line, one is settled enough to point at a tag, the other is still in Preview with behavior that may change; they do not sit on the same maturity scale.

free-claude-code is a third state: a rolling version line. The latest release is v6.19.11 at 2026-11-12 10:04:07Z, and the tags run from v6.19.11 back through v6.19.4, with no 1.0.0-style version anywhere. The repository is Python, created 2026-01-28, with 69898 stars and 10832 forks as cumulative totals. There is no formal 1.0 milestone here; the number is simply a rolling iteration counter. Lining up a Preview, a formal tag and a rolling version on one row and calling it a ranking is meaningless, so this article pins each one to its own timestamp and then looks at where its access boundaries fall.

2. Access-boundary comparison table

The table below compares the three across five dimensions. Each cell holds only first-hand facts that can be checked word for word in the repository README, LICENSE and release notes; where no such statement exists, the cell says so rather than filling in a conclusion.

Dimensionqwen-codeOpenBotfree-claude-code
Models and providersQwen and Qoder accounts, multiple model providers, custom compatible endpointsAny endpoint speaking AG-UI; agents from LangGraph, Mastra, CrewAI, Pydantic AI, Google ADK or hand-written12 harnesses routed to 61 providers, covering free, paid, subscription and local
Protocol and endpointsMCP, Plan Mode, LSP Integration; Computer Use native MCP tools with Node REPL and SDKAG-UI as the agent access protocol, MCP for tools, Kubernetes as the default sandboxPreserved native streaming, tool use and interleaved thinking; harness connected to provider endpoints by configuration
Permission and approvalOfficial documentation does not specify thisSign-in, tool permissions, action policies and audit enforced at the API; every action decided before it happens and recorded afterConfigurable rate limits, concurrency control, multi-API-key fallback and fcc-doctor diagnostics; approval model not documented
Version stabilityv1.0.0-preview is a Preview; interfaces, configuration and behaviors may change during Preview; main line v0.27.3 runs in parallelv1.0.2 is a formal tag, alongside v1.0.0, v0.3.4 and fifteen tags totalv6.19.11 is a rolling version with no 1.0.0-style release
Terms highlightsApache-2.0; README also lists a Terms of Service and Privacy Notice and a Chrome extension data-handling statementMIT; the MIT template can be used as-is, white-labeled or deeply customizedAGPL-3.0-only; the GitHub /license endpoint returns NOASSERTION; README states it is not affiliated with Anthropic and removes integrations under provider TOS; ToS friendly is a self-declaration

A few cells deserve a note. Under models and providers, qwen-code is describing accounts and compatible endpoints, OpenBot is describing an AG-UI endpoint that is not tied to a framework, and free-claude-code is describing a count of routable combinations; the three are not talking about the same object. The version-stability row is where the contrast from the opening becomes concrete: a Preview, a formal tag and a rolling version each carry a different rhythm of change. The license appears here as one comparison dimension among several, not as the main axis of this article, and the terms detail is taken up in section four.

3. The structural differences

qwen-code is a terminal harness that folds in Qoder capabilities. The README says the open-source release brings together the agent harness capabilities of Qoder and Qwen Code, and the Agentic out of the box line lists Auto-Memory, Auto-Skills, SubAgents, Agent Teams, and MCP. The feature table adds MCP, Plan Mode, LSP Integration; multi-client integration covers VS Code, Zed, JetBrains, Telegram, DingTalk, WeChat and Feishu; Computer Use runs on native MCP tools with a Node REPL and SDK. In short, its extension points live inside the harness: MCP attaches tools, SubAgents and Agent Teams split work, and Auto-Memory keeps context. The release notes add native packages for macOS, Linux and Windows with no separate Node.js installation, plus session resumption and non-interactive execution. The repository's open_issues_count is 2257, and the README carries a Terms of Service and Privacy Notice and a Chrome extension data-handling statement; these are boundary documents to read before integrating, but they do not amount to a published approval-layer specification.

OpenBot is an enterprise agent workbench plus governance, not a plain terminal coding tool, and it is not a chat-only product. Its description is Bring any AG-UI agent, the README says From CopilotKit, creators of AG-UI, and states plainly that A Bot is any endpoint speaking AG-UI, the open protocol for agent-to-user interaction, so OpenBot is not tied to a framework. As long as the endpoint speaks AG-UI, an agent built with LangGraph, Mastra, CrewAI, Pydantic AI, Google ADK or by hand arrives the same way. MCP connects tools, AG-UI connects agents, Kubernetes is the default sandbox, and front end and back end can evolve independently. The permission landing point is explicit: OpenBot enforces sign-in, tool permissions, action policies, and audit at its API, while each coworker gets a computer of its own, a browser, files and tools, with every action decided before it happens and recorded after. Its open_issues_count is 107. Approval lives at the API layer here, rather than being scattered across individual tools.

free-claude-code is a routing proxy sitting between harnesses and providers. Its README Features section lists 12 supported harnesses: Claude Code, Codex, Pi, OpenCode, Cline, Hermes, DeepSeek Harness, Grok Build, Muse Code, Aider, Antigravity CLI or VS Code Chat; then 61 supported providers, including free, paid, subscription and local providers; then the summary line Route any supported harness to models from any supported provider, with preserved native streaming, tool use and interleaved thinking. It also lists automatic retries, configurable rate limits, concurrency control, automatic model and provider fallback, multi-API-key fallback and an fcc-doctor diagnostic command. It does not supply models of its own and does not rewrite what a harness can do; its value is in connecting the two sides. The cost is that the terms of both sides land on it at once, which is what the next section covers.

4. What to verify first

If you are wiring qwen-code into a pipeline, check the Preview sentence up front: Some interfaces, configuration options, and behaviors may change during Preview. Both v1.0.0-preview and v1.0.0-preview.1 carry the preview suffix, while the main line is still v0.27.3. If you need a stable interface, do not lock workflows onto the Preview line as if it were a formal 1.0.0 release; leave configuration and automation scripts room to follow changes, and treat the sentence about interfaces, configuration and behaviors changing during Preview as an integration premise rather than a footnote.

If you are evaluating free-claude-code, read the terms column carefully. The root LICENSE identifier line is, verbatim:

text
SPDX-License-Identifier: AGPL-3.0-only

README line 567 says GNU Affero General Public License version 3 only (AGPL-3.0-only). See LICENSE for details. That is -only, not -or-later, and not a loose AGPL-3.0. The GitHub /license endpoint, meanwhile, returns spdx_id=NOASSERTION and key=other, which means GitHub did not auto-detect a standard license; it must not be reported as GitHub having identified it as AGPL. The AGPL network clause is direct: the operator of a network server must provide the source code of the modified version running there to the users of that server. The README states Independent open-source project. Not affiliated with or endorsed by Anthropic, with Claude and Claude Code as Anthropic trademarks; this is not an official Anthropic project. It also says FCC follows provider TOS and removes integrations if they stop being allowed, and Free-tier availability and limits are controlled by each provider and may change. In other words, the 1.3B+ free tokens per month line in the README is the project's own statement about free quota it can route, not a guaranteed available quota; free-tier availability and limits are controlled by each provider and may change. ToS friendly in the description is likewise a self-declaration by the project, not an endorsement from Anthropic or any provider. The provider terms are not set by the project: once an integration is no longer allowed under the provider terms, the project removes it, and any deployment depending on that link should plan for change.

If you are integrating OpenBot, check two landing points: where the AG-UI endpoint and the MCP tools each place permission. The README puts sign-in, tool permissions, action policies and audit at the API, with every action decided before it happens and recorded after; every incoming agent that speaks AG-UI is treated the same, with no framework binding, but governance rules are still applied uniformly by the OpenBot layer. The notarized dmg and the signed Windows installer are distribution details and do not change that permission model.

An Agent runtime and SDK comparison on this site discusses how runtimes are embedded. This article does not decide which project anyone should pick; it lays out version status and access boundaries so readers can check the first-hand files along whichever link they plan to wire up.

FAQ

Q1: Is qwen-code v1.0.0-preview a formal 1.0.0 release? A1: No. It is a Preview. The release notes say verbatim that some interfaces, configuration options, and behaviors may change during Preview, so interfaces, configuration and behaviors may change during Preview; the main line was still v0.27.3, and it cannot be treated as a stable production build.

Q2: Can the star counts be read as single-day momentum? A2: No. 30121, 12126 and 69898 are cumulative totals returned by the GitHub API as of November 14, 2026, not single-day gains; comparing them as daily movement is meaningless.

Q3: What is the precise way to write the free-claude-code license? A3: Write AGPL-3.0-only, which is -only rather than -or-later and not a loose AGPL-3.0; the GitHub /license endpoint returns NOASSERTION, the README states it is not affiliated with Anthropic, and it removes integrations under provider TOS once they are no longer allowed.

Q4: Is 1.3B+ free tokens per month a quota that is guaranteed to be available? A4: No. The README lists it as a self-reported feature, and states that free-tier availability and limits are controlled by each provider and may change; free-tier availability is controlled by each provider and may change, so it is not a guaranteed available quota.

Q5: How does this article differ from the other coding-agent comparisons on the site? A5: Those pieces are about capability and selection; this one compares version status and access boundaries only, with no selection verdict, no scores and no ranking. Permission and sandbox details are covered in separate in-site articles.

Related reading

This article is AI-assisted and human-edited. Last updated: 2026-11-15

FAQ

Is qwen-code v1.0.0-preview a formal 1.0.0 release?
No. It is a Preview. The release notes say verbatim that some interfaces, configuration options, and behaviors may change during Preview, so interfaces, configuration and behaviors may change during Preview; the main line was still v0.27.3, and it cannot be treated as a stable production build.
Can the star counts be read as single-day momentum?
No. 30121, 12126 and 69898 are cumulative totals returned by the GitHub API as of November 14, 2026, not single-day gains; comparing them as daily movement is meaningless.
What is the precise way to write the free-claude-code license?
Write AGPL-3.0-only, which is -only rather than -or-later and not a loose AGPL-3.0; the GitHub /license endpoint returns NOASSERTION, the README states it is not affiliated with Anthropic, and it removes integrations under provider TOS once they are no longer allowed.
Is 1.3B+ free tokens per month a quota that is guaranteed to be available?
No. The README lists it as a self-reported feature, and states that free-tier availability and limits are controlled by each provider and may change; free-tier availability is controlled by each provider and may change, so it is not a guaranteed available quota.
How does this article differ from the other coding-agent comparisons on the site?
Those pieces are about capability and selection; this one compares version status and access boundaries only, with no selection verdict, no scores and no ranking. Permission and sandbox details are covered in separate in-site articles.

Related