GLM-4.5-Flash from Z.ai is the free tier of the 4.5 generation: $0.00 per Mtok input and output on Z.ai's listed pricing, with the full 128K tokens (131,072) context and the 98,304 output cap of the paid GLM-4.5 models. Rate limits are the price. Inside Atlas, GLM-4.5-Flash is the zero-cost way to put a reasoning model behind the small_model slot, which means chat titles, session summaries, and subagent chatter cost you nothing. Throughput is rate limited, so keep a paid model in the main slot.
What does GLM-4.5-Flash cost to run in Atlas?
GLM-4.5-Flash is listed at $0.00 per Mtok input and output on Z.ai's pricing, which makes Atlas background traffic genuinely free. Chat titles, session summaries, and subagent chatter all stop appearing on the bill entirely when GLM-4.5-Flash sits in the small_model slot.
Background traffic is not trivial in Atlas. Atlas fans out work to subagents that can run in the foreground or in parallel background sessions, and every one of those sessions generates tokens that somebody pays for. GLM-4.5-Flash takes that line item to zero. The tradeoff Z.ai makes explicit is throughput: free means rate limited. You are not paying in dollars, you are paying in requests per minute, and that is a very different constraint to plan around than a per-token rate.
Is GLM-4.5-Flash a cut-down model with a small window?
No. GLM-4.5-Flash keeps the full 131,072 token context and the 98,304 output ceiling of the paid GLM-4.5 tier rather than being cut down to a toy window. Z.ai limits GLM-4.5-Flash on throughput, not on capacity.
Free tiers usually punish you on the window, which is what makes them useless for agentic coding. GLM-4.5-Flash does not. A 131,072 token context is enough for Atlas to push a real retrieval result into the prompt, and the 98,304 output cap is enough to emit a substantial edit. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, so what lands in that 131,072 window is dense and relevant rather than padded, which is what makes a free tier of this shape actually productive.
Can GLM-4.5-Flash handle multi-step plans?
Yes. GLM-4.5-Flash has reasoning enabled, so even the free slot can handle a multi-step plan rather than only single-shot completions. Z.ai kept reasoning on across the 4.5 generation, including the $0.00 per Mtok tier.
A reasoning-capable free model changes what you can delegate to the cheap slot. Atlas drafts a plan in a read-only plan agent and asks before switching to a build agent, and Atlas fans out subagents that need to make real decisions rather than just summarize. GLM-4.5-Flash can carry that kind of work. Every tool call it makes is still permission-gated against allow, ask, and deny rules before it runs, so handing a free model more autonomy inside Atlas does not mean handing it unchecked write access.
What happens when GLM-4.5-Flash hits its rate limits?
Free means rate limited, and Atlas's parallel subagents will hit those limits before a paid tier would. Keeping a paid model in the "model" slot is what stops your interactive loop from stalling when GLM-4.5-Flash starts throttling background work.
Parallel fan-out is precisely the traffic shape that trips a free tier. Several Atlas subagents running at once means several concurrent requests, and a rate-limited endpoint responds to that by queuing or refusing. The documented configuration is defensive: put GLM-4.5-Flash in "small_model" so the free tier absorbs background work, and keep a paid model in "model" so the loop you are actually watching never depends on a throttled endpoint. That split gives you the cost benefit without putting your interactive session at the mercy of a free quota.
Should you use GLM-4.7-Flash instead of GLM-4.5-Flash?
For most Atlas users, yes. GLM-4.7-Flash is the newer free option from Z.ai with a larger 200,000 token context, so GLM-4.5-Flash is only the right pick when you specifically need this checkpoint rather than the newest free tier.
Two free models, one with a bigger window, is not a hard comparison. GLM-4.5-Flash holds 131,072 tokens; GLM-4.7-Flash holds 200,000. If nothing pins you to the 4.5 generation, take the larger window. The reasons to stay on GLM-4.5-Flash are specific: an evaluation already validated against this checkpoint, prompt behavior tuned to it, or a deliberate decision to keep the whole small_model path inside the 4.5 family. Atlas lets you switch the active model and provider on the fly with favorites and recents, so testing both against your own workload costs almost nothing.
Setup
- 01Sign up at z.ai, then export ZHIPU_API_KEY or run `atlas login` and pick Z.ai.
- 02Run `atlas models zai` and confirm that `glm-4.5-flash` appears in the list.
- 03Wire GLM-4.5-Flash as the free background slot: set "small_model": "zai/glm-4.5-flash" in atlas.json.
- 04Keep a paid model in "model" so that when GLM-4.5-Flash rate limits, your interactive loop does not stall.
- 05Watch for throttling when Atlas fans out parallel subagents, since concurrency is what trips the free tier's limits first.
Frequently asked questions
- is glm-4.5-flash free to use
- Yes. GLM-4.5-Flash is listed at $0.00 per Mtok input and output on Z.ai's pricing. Throughput is rate limited, which is the real cost.
- what is the glm-4.5-flash context window
- GLM-4.5-Flash keeps the full 128K tokens (131,072) context and 98,304 output cap of the paid GLM-4.5 tier.
- how do I use glm-4.5-flash as atlas's free model
- Sign up at z.ai, export ZHIPU_API_KEY, run `atlas models zai` to confirm `glm-4.5-flash`, then set "small_model": "zai/glm-4.5-flash" in atlas.json.
- does glm-4.5-flash support reasoning
- Yes. GLM-4.5-Flash has reasoning enabled, so it can carry a multi-step plan rather than being limited to single-shot completions.
- will glm-4.5-flash rate limit atlas subagents
- Likely. Atlas's parallel subagents hit free-tier rate limits before a paid tier would, so keep a paid model in the "model" slot.
- glm-4.5-flash vs glm-4.7-flash
- GLM-4.7-Flash is the newer free option with a larger 200,000 context. GLM-4.5-Flash is only right if you specifically need the 4.5 checkpoint.
- can I run atlas entirely on a free model
- You can, but Z.ai rate limits GLM-4.5-Flash. Keeping a paid model in "model" prevents your interactive Atlas loop from stalling under throttling.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas vs Base44: Terminal AI Coding Agents in 2026
Compare Atlas, the terminal-native AI coding agent, with Base44, the Wix-owned no-code app builder, for developers in 2026. Evaluate features, pricing, and workflow.
Atlas for TypeScript in 2026
In 2026, TypeScript developers leverage Atlas, the terminal-native AI coding agent, to enhance productivity. Atlas understands your types, ensures code quality, and offers robust safety features.
Atlas vs Devin: AI Coding Agents Compared for 2026
Atlas and Devin offer distinct AI coding experiences in 2026. Atlas provides a terminal-native TUI with local control, while Devin is a cloud-managed engineer with SWE-1.7.
Atlas vs Amazon Q Developer: Terminal AI Coding Agents in 2026
Comparing Atlas and Amazon Q Developer in 2026. Atlas offers terminal-native AI with permission-gated tools and local code indexing. Amazon Q Developer excels in AWS services, providing Java upgrades and security
Atlas vs Tabnine: Choosing Your Terminal AI Coding Agent in 2026
Comparing Atlas and Tabnine for developers in 2026. Atlas offers a terminal-native AI agent with diff review, while Tabnine provides privacy-first code completion and chat.
Run Atlas Headless in CI with Atlas (2026 Workflow)
How to run Atlas headless in CI in 2026: atlas run sends one prompt and exits when the session goes idle, with --format json, --command, and --continue for pipeline steps.
Trace a Runtime Bug from a Stack Trace with Atlas in 2026
How to trace a runtime bug from a stack trace with Atlas in 2026: read each frame at its offset, grep for the error string, and use the lsp tool to find callers.
Atlas for Phoenix in 2026
Atlas is a terminal-native AI coding agent for Phoenix in 2026. It reads contexts, LiveView modules, and Ecto changesets, then runs mix test behind a prompt.