# Atlas with GLM-4.7-FlashX: The Cheapest Paid Slot in 2026 > GLM-4.7-FlashX costs $0.07 per Mtok input and $0.40 per Mtok output on a 200,000 token context, the cheapest non-free model in the comparison. GLM-4.7-FlashX from Z.ai is the cheapest non-free model in the Atlas model comparison: $0.07 per Mtok input and $0.40 per Mtok output, with a 200K tokens (200,000) context and reasoning enabled. GLM-4.7-FlashX is what you point Atlas's small_model slot at when the free tier's rate limits start biting. At $0.40 per Mtok output, GLM-4.7-FlashX changes an agent loop from something you supervise into something you can simply let run, and the 131,072 output cap means the cheap tier is not restricted to short answers. ## Key takeaways - GLM-4.7-FlashX costs $0.07 per Mtok input and $0.40 per Mtok output, the cheapest non-free model in the comparison. - $0.07 per Mtok input is half of DeepSeek Chat's $0.14 and roughly one ninth of GLM-4.7's $0.60, on the same 200,000 token context. - $0.40 per Mtok output is under a fifth of GLM-4.7's $2.20, which changes an agent loop from supervised to unattended. - Reasoning is enabled and the output cap is a full 131,072 tokens, so the cheap tier is not restricted to short answers. - GLM-4.7-FlashX is a distilled speed tier and loses to GLM-4.7 and GLM-5 on genuinely hard architectural reasoning. ## How cheap is GLM-4.7-FlashX compared to other models? GLM-4.7-FlashX charges $0.07 per Mtok input, which is half of DeepSeek Chat's $0.14 and roughly one ninth of GLM-4.7's $0.60, on the same 200,000 token context. Z.ai prices output at $0.40 per Mtok, under a fifth of GLM-4.7's $2.20. Those two ratios are the entire case for GLM-4.7-FlashX. One ninth the input rate of GLM-4.7 while carrying the same 200,000 token window means Atlas can push a full retrieval result into the prompt without the input side of the bill mattering. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, and that fused result set is large by design. On GLM-4.7-FlashX at $0.07 per Mtok input, filling a 200,000 token window is a cost you stop thinking about. ## What does $0.40 per Mtok output change about an Atlas agent loop? GLM-4.7-FlashX's $0.40 per Mtok output is under a fifth of GLM-4.7's $2.20, which is the difference between an agent loop you watch and one you can just let run. Atlas fans out subagents in parallel background sessions, and output price multiplies across every one of them. Supervision is expensive in a way that never shows on an invoice. If a developer feels obliged to watch an Atlas run because the meter is spinning, the agent is not really saving time. GLM-4.7-FlashX removes that pressure. Long tool-calling loops, wide subagent fan-out, and repeated retries all become affordable at $0.40 per Mtok output. Atlas still permission-gates every tool call against allow, ask, and deny rules before it runs, so letting a cheap model run unattended does not mean letting it write wherever it likes. ## Is GLM-4.7-FlashX limited to short answers? No. GLM-4.7-FlashX has reasoning enabled and a full 131,072 token output cap, so the cheap tier is not restricted to short answers. Z.ai kept both properties intact at the $0.07 per Mtok input price point. The usual failure of a budget tier is that it can read a lot but cannot write much, which makes it useless for refactoring. GLM-4.7-FlashX does not have that shape. A 131,072 token output ceiling is enough for a substantial multi-file diff in a single response, and reasoning being enabled means GLM-4.7-FlashX can plan the edit rather than pattern-match it. Atlas computes a unified diff for every file edit and surfaces it for approval before writing, so a large output from a cheap model still lands in front of you as a reviewable change. ## Where does GLM-4.7-FlashX lose to GLM-4.7 and GLM-5? GLM-4.7-FlashX is a distilled speed tier, so on genuinely hard architectural reasoning it will lose to GLM-4.7 and GLM-5. The $0.07 per Mtok input price reflects a real capability tradeoff, not a promotional discount. Distillation buys throughput and cost by giving up depth on the hardest problems. In Atlas terms, that means GLM-4.7-FlashX is strong on mechanical work, renames, test scaffolding, and repetitive migrations, and weaker when Atlas drafts a plan in a read-only plan agent for a decision that will be expensive to reverse. The practical split is to keep GLM-4.7 or GLM-5 in the "model" slot for planning, and GLM-4.7-FlashX in "small_model" for the volume. Atlas lets you switch the active model and provider on the fly with favorites and recents, so the boundary is not fixed. ## Should you use GLM-4.7-Flash instead of GLM-4.7-FlashX? GLM-4.7-Flash is free at the same generation as GLM-4.7-FlashX, so FlashX only makes sense once you need throughput past the free tier's limits. Start on GLM-4.7-Flash, and move to GLM-4.7-FlashX at $0.07 per Mtok input when rate limits start blocking Atlas subagents. The upgrade path is clean because both models sit in the same generation and require no prompt changes. Run GLM-4.7-Flash in the small_model slot first and see whether Atlas's parallel background sessions stay inside the free quota. When they do not, and free-tier throttling starts stalling subagent work, GLM-4.7-FlashX is a one-line switch to "small_model": "zai/glm-4.7-flashx" in atlas.json. You are paying $0.07 per Mtok input and $0.40 per Mtok output purely to remove a rate limit, which is the cheapest problem in this entire lineup to solve. ## Setup 1. Export ZHIPU_API_KEY, or run `atlas login` and pick Z.ai. 2. Run `atlas models zai` and locate `glm-4.7-flashx` in the list. 3. Set "small_model": "zai/glm-4.7-flashx" in atlas.json so Atlas titles, summaries, and subagents run at $0.07 per Mtok input. 4. Optionally set GLM-4.7-FlashX as the main "model" too, for high-volume mechanical refactors where the cost curve matters more than the last few points of reasoning quality. 5. Keep GLM-4.7 or GLM-5 in the main slot when planning quality matters, since GLM-4.7-FlashX is a distilled speed tier. ## FAQ ### how much does glm-4.7-flashx cost GLM-4.7-FlashX costs $0.07 per Mtok input and $0.40 per Mtok output, making it the cheapest non-free model in this comparison. ### what is the glm-4.7-flashx context window GLM-4.7-FlashX has a 200K tokens (200,000) context window with a 131,072 output cap, and reasoning is enabled. ### is glm-4.7-flashx cheaper than deepseek chat Yes on input. GLM-4.7-FlashX charges $0.07 per Mtok input, half of DeepSeek Chat's $0.14, on a 200,000 token context. ### how do I set glm-4.7-flashx as atlas's small model Export ZHIPU_API_KEY, run `atlas models zai` to locate `glm-4.7-flashx`, then set "small_model": "zai/glm-4.7-flashx" in atlas.json. ### can I use glm-4.7-flashx as my main atlas model Yes, for high-volume mechanical refactors where the cost curve matters more than the last few points of reasoning quality. ### glm-4.7-flashx vs glm-4.7-flash GLM-4.7-Flash is free at the same generation. GLM-4.7-FlashX at $0.07 / $0.40 only makes sense once you need throughput past the free tier's limits. ### is glm-4.7-flashx good at architectural reasoning No. GLM-4.7-FlashX is a distilled speed tier and will lose to GLM-4.7 and GLM-5 on genuinely hard architectural reasoning. --- Canonical HTML: https://runatlas.sh/resources/models/glm-4-7-flashx Source of truth: aeo_pages row `/resources/models/glm-4-7-flashx` (segment: Models) (this file is generated from it, never hand-edited). Licence: Atlas is proprietary with a free core. It is not open source and there is no public source repository.