Open-Source Claude Code Alternative (2026, Free)
Identifying Claude Code · Mark source
This hub lists open-source coding agents people compare with Claude Code-class terminal and IDE agents. Claude Code is named only to identify that product class. We copy licenses and GitHub star counts from official repositories. We do not rank agent quality against Claude Code.
- Cline — Open-source coding agent for VS Code, JetBrains, CLI, SDK, and a desktop app.
- OpenCursor — MIT VS Code extension for agentic chat, local and cloud LLMs, semantic search, and MCP.
- Headroom — Apache-2.0 context compression layer for AI agents (library, proxy, MCP).
How we pick
- GitHub repository we could open, with a push in the 90 days before 2026-10-19 unless noted.
- License copied from that repository LICENSE file.
- Agentic coding in a terminal, IDE, CLI, or desktop app.
- No unsourced "beats Claude Code" claims.
Comparison
| Name | License | Price | Platforms | Runs locally | Key strengths |
|---|---|---|---|---|---|
| Cline | Apache-2.0 | free | VS Code, JetBrains, CLI, macOS, Windows | Yes | Open-source coding agent for VS Code, JetBrains, CLI, SDK, and a desktop app. |
| OpenCursor | MIT | free | VS Code | Yes | MIT VS Code extension for agentic chat, local and cloud LLMs, semantic search, and MCP. |
| Headroom | Apache-2.0 | free | Linux, macOS, Windows, Python, Node | Yes | Apache-2.0 context compression layer for AI agents (library, proxy, MCP). |
| Orca | MIT | free | macOS, Windows, Linux | Yes | MIT desktop orchestrator for running coding agents side-by-side in worktrees. |
| DeepSeek Harness | MIT | free | Linux, macOS, Windows, Node.js, Web | Yes | MIT plugin-based agent harness from DeepSeek. Developer preview; breaking changes expected. |
| Hindsight | MIT | free | Linux, macOS, Windows, Docker, Kubernetes, Python, Node.js, Go | Yes | MIT agent memory that learns — retain, recall, reflect, and banks. Not a chat model. |
| Ling-3.1-flash | MIT | free | See model card | Yes | inclusionAI MIT 560B/25B hybrid-linear MoE with official SGLang and vLLM recipes. |
| Ornith-1.5-122B | MIT | free | See model card | Yes | MIT 122B MoE coding/agent model with 262k context and OpenAI-style tool calling. |
| Flint | Apache-2.0 | free | See model card | Yes | Recursive Apache-2.0 39.8B/2.5B local coding-agent MoE with GGUF, INT4, and Ollama tags. |
| Nanbeige4.3-33B | Apache-2.0 | free | See model card | Yes | Apache-2.0 33B text agent teacher from Nanbeige (BOSS Zhipin) with 512K context. |
| Spark-X2.6-4B | Apache-2.0 | free | See model card | Yes | XHToken Apache-2.0 ~4B on-device agent/coding LLM. Not Meta Muse Spark Open. |
| SenseNova 6.8 Code2Video | Apache-2.0 | free | See model card | Yes | SenseTime Apache-2.0 27B code-driven video (Qwen3.8-27B fine-tune). Not a pixel DiT. |
| Mistral Large 4 | other | free | See model card | Yes | Mistral ~1T/52B ungated multimodal MoE (text+image→text), 1M context. Le Chonk. |
| Mage-VL-Cowork | Apache-2.0 | free | See model card | Yes | Microsoft Apache-2.0 ~4.7B hybrid computer-use VLM (GUI + bash). Not T2I. |
Lead rows: Cline and OpenCursor. Compare them on Cline vs OpenCursor. Cursor-class editors also live on Cursor alternatives. Copilot-class lists hang off GitHub Copilot alternatives. Agent memory (not an agent): Hindsight on Mem0 alternatives.
Sourced snapshot (2026-10-19)
| Field | Cline | OpenCursor |
|---|---|---|
| Repository | github.com/cline/cline | github.com/PawanOsman/OpenCursor |
| Stars / forks | 70,521 / 7,705 | 6,033 / 1,030 |
| License | Apache-2.0 | MIT |
| Created | 2024-07-06 | 2022-12-06 |
| Last push | 2026-10-19T01:46:35Z | 2026-09-25T08:24:32Z |
| Type | Agent (VS Code, CLI, desktop, SDK) | VS Code extension |
| Quality vs Claude Code | Not independently tested | Not independently tested |
Why these two
Claude Code is Anthropic’s coding agent for terminal, IDE, Slack, and web (product page, opened 2026-10-19). It is listed on Pro, Max, Team, and Enterprise plans. Searchers type “open source Claude Code alternative” when they want an auditable agent they can point at local or third-party models.
Cline is the highest-star actively pushed agent we opened today. OpenCursor is the lightest VS Code-only option still inside the 90-day window.
Models attached to this hub
These are weights, not Claude Code clones. Ling-3.1-flash is inclusionAI’s ungated MIT 560B/25B hybrid MoE (121 Hugging Face likes); its README names Claude Code among harnesses it “delivers a strong user experience” in — that sentence is vendor copy, not a forkalt test. Reflection Beam is an ungated Apache-2.0 501.3B/23.4B text MoE with Mirror CLI. Ornith-1.5-122B is an ungated MIT 122B MoE whose card publishes vLLM/SGLang tool-calling recipes (129 likes, 2026-10-22); compare Ornith vs DeepSeek-V4.1-Flash. Flint is Recursive’s ungated Apache-2.0 39.8B/2.5B local coding MoE (533 likes, 2026-10-23) with INT4/GGUF siblings and an Ollama library tag that names ollama launch claude --model flint — that launcher label is upstream, not a forkalt test; compare Flint vs Ornith-1.5-122B. Nanbeige4.3-33B is Nanbeige / BOSS Zhipin’s ungated 33B text teacher (365 likes, 2026-10-26) whose card targets coding and professional cowork; compare Nanbeige4.3-33B vs Qwen3.8-27B. Spark-X2.6-4B is XHToken’s ungated Apache-2.0 ~4B on-device agent/coding LLM (514 likes, 2026-10-26) — not Meta Muse Spark Open; compare Spark-X2.6-4B vs LFM2.5-24B-A2B. SenseNova 6.8 Code2Video is SenseTime’s ungated 27B code-driven video checkpoint (45 likes, 2026-10-29); the README’s recommended harness is Claude Code pointed at a self-hosted endpoint. That is a video job, not a Claude Code clone. Mistral Large 4 is Mistral’s ungated ~1T/52B multimodal MoE (1,301 likes, 2026-10-30); the card names coding/agent recipes, vLLM/SGLang parsers, and Mistral Vibe — those labels are upstream, not a forkalt test; compare vs Muse Spark Open. Mage-VL-Cowork is Microsoft’s ungated Apache-2.0 ~4.7B hybrid computer-use VLM (93 likes, 2026-11-01) with GUI and bash channels — not a Claude Code clone and not Iris-3B.
How to choose
- Need CLI + desktop + VS Code: Cline.
- Need a VS Code extension only: OpenCursor.
- Choosing between those two: Cline vs OpenCursor.
- Need a Cursor-class IDE hub: Cursor alternatives.
- Need MIT datacenter weights the vendor names next to Claude Code: Ling-3.1-flash.
- Need a MIT 122B MoE with a published vLLM/SGLang recipe: Ornith-1.5-122B.
- Need Apache-2.0 39.8B/2.5B local coding weights with GGUF/Ollama: Flint.
- Need a 33B Apache-2.0 text teacher (512K, agent/coding card): Nanbeige4.3-33B.
- Need Apache-2.0 ~4B on-device coding/agent weights (Ollama named): Spark-X2.6-4B.
- Need a 27B code-to-video checkpoint whose README names Claude Code as the harness: SenseNova 6.8 Code2Video.
- Need Mistral’s ~1T/52B open-weight coding/agent MoE (Vibe named on the card): Mistral Large 4.
- Need Microsoft’s Apache-2.0 ~4.7B hybrid CUA (screenshot GUI + bash): Mage-VL-Cowork.
- Need a context-compression layer (not an agent): Headroom (Apache-2.0, trending 2026-10-21).
- Need a fleet/worktree orchestrator that can run Claude Code beside others: Orca (MIT, trending 2026-10-21).
- Need DeepSeek’s MIT plugin harness (
dsh, developer preview): DeepSeek Harness (260,528 stars, dsh-v0.2.1-rc.3, 2026-10-25). - Need a MIT agent memory layer (retain/recall/reflect; not an agent): Hindsight (56,654 stars, v1.1.0, 2026-10-31) on Mem0 alternatives.
See also ChatGPT alternatives and Dev & Coding.
Listed options
Cline
Open-source coding agent for VS Code, JetBrains, CLI, SDK, and a desktop app.
OpenCursor
MIT VS Code extension for agentic chat, local and cloud LLMs, semantic search, and MCP.
Headroom
Apache-2.0 context compression layer for AI agents (library, proxy, MCP).
Orca
MIT desktop orchestrator for running coding agents side-by-side in worktrees.
DeepSeek Harness
MIT plugin-based agent harness from DeepSeek. Developer preview; breaking changes expected.
Hindsight
MIT agent memory that learns — retain, recall, reflect, and banks. Not a chat model.
Ling-3.1-flash
inclusionAI MIT 560B/25B hybrid-linear MoE with official SGLang and vLLM recipes.
Ornith-1.5-122B
MIT 122B MoE coding/agent model with 262k context and OpenAI-style tool calling.
Flint
Recursive Apache-2.0 39.8B/2.5B local coding-agent MoE with GGUF, INT4, and Ollama tags.
Nanbeige4.3-33B
Apache-2.0 33B text agent teacher from Nanbeige (BOSS Zhipin) with 512K context.
Spark-X2.6-4B
XHToken Apache-2.0 ~4B on-device agent/coding LLM. Not Meta Muse Spark Open.
SenseNova 6.8 Code2Video
SenseTime Apache-2.0 27B code-driven video (Qwen3.8-27B fine-tune). Not a pixel DiT.
Mistral Large 4
Mistral ~1T/52B ungated multimodal MoE (text+image→text), 1M context. Le Chonk.
Mage-VL-Cowork
Microsoft Apache-2.0 ~4.7B hybrid computer-use VLM (GUI + bash). Not T2I.
FAQ
Is there a free open-source Claude Code alternative I can use?
Two lead rows as of 2026-10-19. Cline (cline/cline) is Apache-2.0 with 70,521 GitHub stars, last push 2026-10-19: an agent for VS Code, CLI, and desktop. OpenCursor (PawanOsman/OpenCursor) is MIT with 6,033 stars, last push 2026-09-25: a VS Code extension with local Ollama/llama.cpp. "Open-source" means the agent code is downloadable; cloud model APIs you attach may still be paid.
Does it match Claude Code quality?
We have no sourced bake-off. Agent quality depends on the model you attach. Claude Code is a hosted Anthropic product on Pro/Max/Team/Enterprise or Console API billing. That is product class, not a ranking.
What does Claude Code cost?
Anthropic's pricing page (opened 2026-10-19) lists Claude Code on Pro ($20/month, or $17/month billed annually), Max (from $100/month), Team, and Enterprise. A Claude Console account bills API tokens. Those figures identify the paid product; they are not a quality column.
Which open agent should I try first?
Cline if you want CLI plus desktop. OpenCursor if you want a VS Code extension only. See Cline vs OpenCursor. Not a quality rank.
Sources & verification
- https://github.com/cline/cline
- https://github.com/PawanOsman/OpenCursor
- https://claude.com/product/claude-code
- https://claude.com/pricing?mode=code
- https://huggingface.co/inclusionAI/Ling-3.1-flash
- https://huggingface.co/reflection/Beam-1.0-501B-A23B
- https://huggingface.co/ornith-ai/Ornith-1.5-122B
- https://huggingface.co/RecursiveSI/flint
- https://huggingface.co/Nanbeige/Nanbeige4.3-33B
- https://huggingface.co/XHToken/Spark-X2.6-4B
- https://huggingface.co/sensenova/SenseNova-6.8-Code2Video
- https://github.com/OpenSenseNova/SenseNova-6.8-Code2Video
- https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B
- https://mistral.ai/news/mistral-large-4-oss/
- https://github.com/vectorize-io/hindsight
- https://github.com/vectorize-io/hindsight/blob/main/LICENSE
- https://huggingface.co/microsoft/Mage-VL-Cowork
- https://github.com/microsoft/Mage
License notes are not legal advice. Read the original LICENSE. Trademark & disclaimer.