The open-weight coding starter stack
For developers who want to cut model cost without leaving the client they already use.
Keep your current editor and change what feeds it. The GLM Coding Plan runs open-weight models inside tools like Claude Code, so routine work moves to a flat monthly subscription while a frontier seat stays for the hard fraction. The honest shape is a daily driver plus a retained frontier model, not a replacement: on ambiguous specs and sustained multi-agent work the closed frontier models still win. One governance note to weigh before you route real code through it: GLM's hosted API is served from China.
The stack, step by step
- 01
GLM (Z.ai)
The daily driver: the GLM Coding Plan runs GLM's open-weight models inside your existing client for a flat monthly fee, so routine coding stops metering frontier tokens.
- 02
Claude Code
The retained frontier seat: kept for ambiguous specs and multi-file reasoning where the closed model still leads, used sparingly rather than for everything.
- 03
GitHub Copilot
The baseline: free-tier completions and GitHub-native chat underneath, so inline help costs nothing while the agents take the heavier work.
What it costs
A flat GLM Coding Plan subscription carries the routine load, Copilot's free tier covers completions, and the frontier seat is metered only for the hard fraction, so the bill is a small fixed plan plus occasional frontier usage rather than frontier tokens all day.
| Tool | Entry tier | What drives cost up |
|---|---|---|
| GLM (Z.ai) | Free tier + paid plans | The GLM Coding Plan runs from around $18/mo as a flat subscription that feeds tools such as Claude Code, with pay-per-token API access alongside; GLM-5.2's MIT-licensed open weights cost nothing to download and run yourself, with spend shifting to your own hardware. |
| Claude Code | Paid only | Included with Claude Pro ($20/mo), Max ($100/$200/mo) or Team Premium; or pay per token via the API (Opus 4.8 $5/$25 per million input/output tokens). |
| GitHub Copilot | Free tier + paid plans | Free ($0); Pro ($10/mo, includes $15/1,500 credits); Pro+ ($39/mo, 7,000 credits); Business ($19/user/mo); Enterprise ($39/user/mo). 1 AI Credit = $0.01. |
Compare the members
Written comparisons between these tools and their nearest substitutes.
The three tiers of this stack
Ready · this stack
The open-weight coding starter stack
For developers who want to cut model cost without leaving the client they already use.
Competitive
The open-weight coding agent stack
For teams that want a dedicated open-weight agent as the daily driver, with a frontier seat kept for the hard fraction.
World-Class
The self-hosted coding stack
For teams whose code cannot leave their own infrastructure, running open weights they control end to end.
Common questions
What does this stack actually cost per month?
Two of the three tools have a free tier to start; Claude Code is paid from day one. The 03 COSTS table above breaks down each tool's pricing. Two things move the bill: the frontier seat, metered by how much of the hard fraction you route to Claude Code, and Copilot's credits once you work past its completions. The GLM plan sits flat underneath and does not climb with use. So the variable to watch is how often you reach past the daily driver, not the subscription.
Do I need all three tools from day one?
Rarely. The GLM Coding Plan is the piece that matters on day one: it slots the open-weight models into the client you already run and moves routine work off frontier tokens. The numbered steps rank the tools by role, not the order you adopt them. Keep the frontier seat, Claude Code, for the ambiguous specs GLM stalls on; Copilot's free completions can sit underneath at no added cost.
I already use GitHub Copilot. What changes?
Your Copilot subscription already gives you inline completions and GitHub-native chat, which earn their keep: leave them on. What it lacks is a cheap daily driver for routine volume and a frontier seat for the hard fraction. This stack adds both: the GLM Coding Plan running open-weight models inside your client at a flat fee, and Claude Code for the ambiguous, multi-file specs where the frontier leads.
Where do these tools overlap, and which wins?
Claude Code and GitHub Copilot both put AI in your editor, the one real overlap. The dividing rule is scope. Copilot wins at the keystroke: inline completions and GitHub-native chat. Claude Code wins on the harder unit: ambiguous specs, multi-file changes, sustained agentic runs. Reach for it when a task needs reasoning across files, not lines finished. The Claude Code and GitHub Copilot comparison, linked above, covers the finer calls.
When do I outgrow this stack?
Two signs. The first: the hard fraction stops being a fraction, and you route more work to the frontier seat than the flat plan, so the daily driver is no longer carrying the load. The second: you want open-weight models running longer, orchestrated agent loops rather than assisted sessions inside your editor. Both point to the competitive tier, the open-weight coding agent stack.
What can I safely put into these tools?
Treat the defaults as leaky. Claude Code's consumer plans train on what you send unless you opt out, and Copilot excludes your code from training only on its Business and Enterprise tiers: route proprietary code through the commercial or training-excluded plans, not the individual defaults. GLM adds a jurisdiction question, since its hosted API is served from China; for sensitive code, weigh that or run the open weights yourself.
Before sharing confidential or personal data, check this tool's data-governance and training policies. They differ between providers and can change.
Last checked: July 2026
Where to start
Not sure what to adopt first?
Five quick questions about your job, task and constraints. We'll suggest your top three tools, plus the one to try first.
Tool facts last checked July 2026