Z.ai · Available now
GLM 5.3 Flash (2026): Z.ai’s low-cost multimodal model, its price, specs and release date, and how it differs from FlashX and GLM-5.3
GLM 5.3 Flash is Z.ai’s low-cost, natively multimodal model with 320B total and 18B active parameters and a 1M-token context, listed in Z.ai’s release notes on August 26, 2026 at $0.15 / $0.50 per million input / output tokens, with MIT-licensed weights on Hugging Face and the same list price on OpenRouter.
Use it now
Chat
≈ 1 credit per reply · See all models
New accounts get 5 free credits once, valid for 7 days. They work on the lower-priced models here; the priciest ones need credits you bought.
Start chatting — 5 free credits- Maker
- Z.ai
- Type
- Text
- Released
- 2026-08-26
- OpenRouter id
- z-ai/glm-5.3-flash
Per million tokens
Price
| - | Input | Output | Source |
|---|---|---|---|
| Z.ai — GLM-5.3-Flash on Z.ai’s own API | $0.15 | $0.50 | docs.z.ai |
| Via OpenRouter (z-ai/glm-5.3-flash) | $0.15 | $0.50 | OpenRouter list row for z-ai/glm-5.3-flash; the 42 endpoints behind it are priced from $0.0399 / $0.0784 to $0.225 / $0.75 · openrouter.ai |
Checked
Facts, with sources
- Z.ai’s release notes ("New Released") list GLM-5.3-Flash under 2026-08-26, after GLM-5.3 (2026-08-18); it is the newest entry on the page we read. The Hugging Face repository zai-org/GLM-5.3-Flash was created on 2026-08-25 according to Hugging Face’s API, a day before; the model card itself prints no date.Source: docs.z.ai · checked 2026-11-19
- Z.ai’s GLM-5.3-Flash page calls GLM-5.3-Flash/FlashX the first native multimodal model in the GLM-5 series. GLM-5.3-Flash has 320B total parameters with 18B activated and combines sparse and linear attention. Input is video, image, text and file, output is text, context length is 1M and maximum output is 128K tokens. The API model code is glm-5.3-flash.Source: docs.z.ai · checked 2026-11-19
- Z.ai’s pricing page lists GLM-5.3-Flash at $0.15 per million input tokens, $0.03 per million cached input tokens and $0.50 per million output tokens, with cached-input storage marked "Limited-time Free". On the same page GLM-5.3-FlashX is $0.37 / $0.075 / $1.25 and GLM-5.3 is $1.4 / $0.26 / $4.4.Source: docs.z.ai · checked 2026-11-19
- Z.ai’s page says thinking cannot be disabled on GLM-5.3-Flash (thinking.type supports only enabled). It recommends temperature 1, top_p 0.95 and reasoning_effort max, and suggests turning on stream and tool_stream for streaming requests.Source: docs.z.ai · checked 2026-11-19
- The Hugging Face model card says the thinking budget is controlled through reasoning_effort with three levels, low, high and max, and that it defaults to max when the parameter is not passed; to use low or high you pass them explicitly. For chat use the card says to pass clear_thinking=true, because the chat template defaults it to false.Source: huggingface.co · checked 2026-11-19
- Z.ai says GLM-5.3-Flash scores 57 on the Artificial Analysis Intelligence Index v4.1.1 at $0.045 per task (discounted), beats GLM-5.2 on six coding and agentic benchmarks (63.4 vs 46.2 on DeepSWE v1.1, 48.8 vs 26.2 on AutomationBench) while approaching Claude Opus 4.8 overall, and reaches 29.0 against Opus 4.8’s 29.5 on its Z.ai Code Bench v1.0 at max effort. These are Z.ai’s own figures; we have not tested them.Source: docs.z.ai · checked 2026-11-19
- The model card says GLM-5.3-Flash outperforms GLM-5.2 across benchmarks and real-world workloads at one-tenth the price, while approaching Claude Opus 4.8 on coding and agentic benchmarks. It adds that the model starts from a newly trained base, uses Manifold-Constrained Hyper-Connections (mHC) and was pre-trained on a 30T-token multimodal corpus. Again these are the maker’s statements, not our results.Source: huggingface.co · checked 2026-11-19
- Z.ai’s page compares the design with GLM-5.3: 320B vs 355B total parameters, 18B vs 32B activated, 45 vs 92 layers, and attention computation cut by 3.01× and KV cache by 4.44×. It also says the KV cache is still slightly larger than that of Kimi-K3 and DeepSeek-V4-Flash. These are the maker’s own figures.Source: docs.z.ai · checked 2026-11-19
- Z.ai lists use cases for GLM-5.3-Flash: coding with visual feedback (turning screenshots or recordings into front-end projects), office deliverables (PPTX, PDF, DOCX, XLSX), financial research, video understanding and editing, Blender 3D scenes, game prototypes in Godot, computer use on real interfaces and CAD-style models. It says that before release it tested the model anonymously as "ox-alpha" on OpenCode and OpenRouter, and that it served it on Chinese AI chips.Source: docs.z.ai · checked 2026-11-19
- Z.ai’s page says GLM-5.3-Flash is available on the GLM Coding Plan with 3× the quota of GLM-5.3 (FlashX is not yet on the plan), and that calls made off-peak, including all day on weekends, use 50% of the standard points.Source: docs.z.ai · checked 2026-11-19
- The Hugging Face repository zai-org/GLM-5.3-Flash is public and not gated; its metadata gives the MIT licence, the pipeline tag image-text-to-text and 321,323,031,390 stored parameters in 62 weight files. The card lists SGLang, vLLM, TokenSpeed, Transformers, KTransformers and Unsloth as ways to run it locally; we did not check hardware needs.Source: huggingface.co · checked 2026-11-19
- OpenRouter’s public model list (November 19, 2026) has z-ai/glm-5.3-flash ("Z.ai: GLM 5.3 Flash") at $0.15 / $0.50 per million input / output tokens, $0.03 for cached input, a 1,048,576-token context, 943,717 maximum completion tokens, text, image and video input, and mandatory reasoning with efforts max, high and low.Source: openrouter.ai · checked 2026-11-19 · second-hand
- OpenRouter’s endpoints list for the model (November 19, 2026) shows 42 endpoints from 36 hosts, priced from $0.0399 / $0.0784 (Relace, fp4, 131,072 maximum completion) up to $0.225 / $0.75 (Fireworks), with Z.AI’s own endpoint at $0.15 / $0.50 (fp8, 131,072 maximum completion).Source: openrouter.ai · checked 2026-11-19 · second-hand
- OpenRouter’s model page for z-ai/glm-5.3-flash shows "Released Aug 26, 2026" and a headline price of $0.0399 / $0.0784 per million tokens, the lowest of its hosts’ prices, not the $0.15 / $0.50 of the list row and of Z.ai’s own endpoint.Source: openrouter.ai · checked 2026-11-19 · second-hand
- On the same OpenRouter list the alias ~z-ai/glm-flash-latest has the alias target z-ai/glm-5.3-flash and is listed at $0.0399 / $0.0784, a different price from the target’s own row; the alias has its own page here, GLM Flash Latest.Source: openrouter.ai · checked 2026-11-19 · second-hand
The GLM-5.3 line
GLM 5.3 Flash next to FlashX, FlashX Legal and GLM 5.3
- Z.ai’s page describes GLM-5.3-FlashX as the version with 200 tokens/s inference speed and GLM-5.3-FlashX-Legal (GLM-Legal) as a model optimised for legal workflows such as legal research, contract review and drafting. For the legal model it says its all-criteria pass rate on Harvey LAB is 19.2%, 12.5 percentage points above the base model; this is Z.ai’s own figure.Source: docs.z.ai · checked 2026-11-19
- OpenRouter’s model list (November 19, 2026) has z-ai/glm-5.3-flashx (created September 18, 2026) and z-ai/glm-5.3-flashx-legal (created November 2, 2026), both at $0.37 / $1.25 per million tokens with a 1,048,576-token context and 131,072 maximum completion tokens. On the same list z-ai/glm-5.3 is at $0.0699 / $1.4 with text-only input and 943,718 maximum completion tokens.Source: openrouter.ai · checked 2026-11-19 · second-hand
- The repository of Sage S2.1 Flash, a separate scientific model from Shanghai Artificial Intelligence Laboratory, says it is built on the GLM-5.3-Flash architecture (see Sage S2.1 Flash). That is a different model from GLM 5.3 Flash.Source: huggingface.co · checked 2026-11-19 · second-hand
Side by side
GLM-5.3-Flash, FlashX, FlashX-Legal and GLM-5.3 compared
Z.ai sells four models in this line at different prices, and OpenRouter lists all four under their own ids. GLM 5.3 Flash is the cheapest on Z.ai’s own price page: $0.15 / $0.50 per million tokens against $0.37 / $1.25 for the faster FlashX (and the legal FlashX-Legal at the same price) and $1.4 / $4.4 for the full GLM-5.3. Only the first three take image and video input on the sources we read; GLM-5.3 is text-only. The table keeps to what the pages state, and cells we could not source are left out. For the full model see GLM 5.3; its high-speed variant is GLM 5.3 Prime, and the OpenRouter pointer to the newest Flash model is GLM Flash Latest. Another low-priced multimodal model is Hy4 Flash, and the next Z.ai number people search for is covered on GLM-5.5. On this site a typical reply to GLM 5.3 Flash costs about 1 credit (a credit is about $0.01 in the $5 pack), billed by its real token cost, at least 1 credit.
| What | GLM-5.3-Flash | GLM-5.3-FlashX | GLM-5.3-FlashX-Legal | GLM-5.3 |
|---|---|---|---|---|
| Z.ai price per million tokens, input / output | $0.15 / $0.50 | $0.37 / $1.25 | $0.37 / $1.25 | $1.4 / $4.4 |
| Z.ai cached input price | $0.03 | $0.075 | $0.075 | $0.26 |
| OpenRouter list price, input / output | $0.15 / $0.50 | $0.37 / $1.25 | $0.37 / $1.25 | $0.0699 / $1.4 |
| Z.ai model code | glm-5.3-flash | glm-5.3-flashx | glm-5.3-flashx-legal | Not stated on the pages we read |
| Context window | 1M tokens | 1,048,576 tokens (OpenRouter) | 1,048,576 tokens (OpenRouter) | 1M tokens |
| Input | Video, image, text and file | Text, image and video (OpenRouter) | Text, image and video (OpenRouter) | Text only |
| Maximum output | 128K tokens | 131,072 tokens (OpenRouter) | 131,072 tokens (OpenRouter) | 128K tokens |
| What Z.ai says it is | First native multimodal GLM-5 model; 320B total, 18B active | The faster version, 200 tokens/s | Optimised for legal workflows | Z.ai’s latest flagship model |
| Date in Z.ai’s release notes | 2026-08-26 | No separate entry | No separate entry | 2026-08-18 |
Source: docs.z.aiSource: openrouter.aiSource: docs.z.aiSource: docs.z.aiSource: docs.z.ai
Nearest available
Models you can use now instead
GLM 5.3
Z.ai$0.08 / $3.6 (per million tokens)
- A secondary report (DataNorth) says Z.ai released GLM-5.3 on August 14, 2026; Z.ai’s docs page calls it "Z.ai’s latest flagship model" but states no date.Source: datanorth.ai · checked 2026-11-15 · second-hand
- Z.ai’s docs say GLM-5.3 advances complex software engineering and agent capabilities, with an emphasis on coding and cybersecurity performance.Source: docs.z.ai · checked 2026-11-15
- Context window 1M tokens, maximum output 128K tokens; text-only input.Source: docs.z.ai · checked 2026-11-15
Hy4 Flash
Tencent$0.04 / $0.14 (per million tokens)
- Tencent’s Hy4 flash announcement, dated November 16, 2026, presents it as the newest member of the Hy model family. Tencent recommends it for most everyday tasks (deep research, document processing, PowerPoint generation, simple front-end work, computer use) and points complex tasks such as long-horizon software engineering to Hy4. Tencent Cloud’s API guide calls it "a leaner, more efficient everyday workhorse".Source: tencent.com · checked 2026-11-17
- The Tencent announcement is dated November 16, 2026, and Tencent Cloud’s model, pricing and API pages for hy4-flash were last updated the same day. The announcement says only that Hy4 flash is available now; it gives no separate release date or version name.Source: tencent.com · checked 2026-11-17
- Tencent says Hy4 flash has 224B total parameters (including 64B Engram parameters), 12B active parameters, a context of up to 1M tokens and multimodal input; the announcement does not state a maximum output or the output types. It says the KV cache is 1/6 the size of Hy4’s and that decoding exceeds 150 tokens per second. These are the maker’s own figures, not tested by us.Source: tencent.com · checked 2026-11-17
DeepSeek V4.1 Flash
DeepSeek$0.30 / $1.2 (per million tokens)
- DeepSeek released V4.1-Flash on September 10, 2026.Source: api-docs.deepseek.com · checked 2026-11-11
- Only the Flash version of V4.1 exists: DeepSeek has published no V4.1 Pro, and the Pro model on its price page is still deepseek-v4-pro (V4-Pro-0813).Source: api-docs.deepseek.com · checked 2026-11-11
- DeepSeek’s pricing page says the API model deepseek-flash is now served by the V4.1-Flash-1017 snapshot.Source: api-docs.deepseek.com · checked 2026-11-11
FAQ
Questions
What is GLM 5.3 Flash?
GLM-5.3-Flash is a model from Z.ai that it calls the first native multimodal model in the GLM-5 series. It has 320B total parameters with 18B activated, mixes sparse and linear attention, takes video, image, text and file input and returns text. Z.ai positions it as a low-cost model for coding, office documents and financial research; its API model code is glm-5.3-flash.
How much does GLM 5.3 Flash cost?
Z.ai’s own API charges $0.15 per million input tokens, $0.03 per million cached input tokens and $0.50 per million output tokens. OpenRouter’s list shows the same $0.15 / $0.50, although its hosts range from $0.0399 / $0.0784 to $0.225 / $0.75. On this site a typical reply costs about 1 credit (a credit is about $0.01 in the $5 pack), billed by its real token cost, at least 1 credit.
When was GLM 5.3 Flash released?
Z.ai’s release notes list GLM-5.3-Flash under 2026-08-26, and OpenRouter shows August 26, 2026 as well. Hugging Face’s API says the model repository was created a day earlier, on 2026-08-25, and Z.ai says it tested the model anonymously as "ox-alpha" before release. The notes list GLM-5.3 itself under 2026-08-18.
What are the specs of GLM 5.3 Flash, and can thinking be turned off?
Z.ai lists a 1M-token context, a maximum output of 128K tokens, and video, image, text and file input. On OpenRouter the context is 1,048,576 tokens with text, image and video input. Thinking cannot be turned off, but the Hugging Face card says reasoning_effort accepts low, high and max, and defaults to max when not passed.
How does GLM 5.3 Flash differ from FlashX, FlashX Legal and GLM-5.3?
FlashX is the faster version (Z.ai says 200 tokens/s) at $0.37 / $1.25, FlashX-Legal is tuned for legal work at the same price, and the full GLM 5.3 costs $1.4 / $4.4 on Z.ai and is text-only. GLM 5.3 Prime is a separate high-speed GLM-5.3 variant on OpenRouter. The table on this page sets them side by side with sources.
Can I download and run GLM 5.3 Flash myself?
The weights are public on Hugging Face as zai-org/GLM-5.3-Flash, not gated, with the MIT licence in the repository metadata. The card lists SGLang, vLLM, TokenSpeed, Transformers, KTransformers and Unsloth as options. With 320B parameters stored it is a large model, and we did not check the hardware it needs, so read the card before you plan a setup. GLM Flash Latest is the hosted alias on OpenRouter.
How good is GLM 5.3 Flash?
By Z.ai’s own figures it scores 57 on the Artificial Analysis Intelligence Index v4.1.1, beats GLM-5.2 on six coding and agentic benchmarks (63.4 vs 46.2 on DeepSWE v1.1) and reaches 29.0 against Claude Opus 4.8’s 29.5 on Z.ai Code Bench at max effort. We have not tested any of this, and Z.ai itself says the model’s KV cache is still larger than that of two rivals.
Is this the official Z.ai website?
No. This page is unofficial and not affiliated with or endorsed by Z.ai. Facts link to the pages we read.