GLM-5.3: Доступность в API
ПодтвержденоОткрытые веса18 ссылок
Оригинал на английском

Z.ai
USD / 1 млн токенов
| Вход | Выход | Кэшированный вход |
|---|---|---|
|
|
|
| Модель | Ссылки на источники | Цена входа / выхода |
|---|---|---|
glm-5.3standard · текст | opencode-go | $1,40 / $4,40 |
glm-5.3standard | zai-pricing Маршрут API недоступен | $1,40 / $4,40 |
Model: glm-5.3
model, the task passed 5 of 5 on GLM-5.3 and 5 of 5 on GLM-5.3-Flash, compared with 4 of 5 and 2 of 5 before.
Thêm Titan 4 Mega (GLM-5.3 Flash) làm model mạnh nhất, mặc định
Add Titan 4 Mega (GLM-5.3 Flash) as the new default and strongest model
- GLM by Z.ai, model id `z-ai/glm-5.3`, run through opencode by the
Claude SDK が --model glm-5.3 フラグを検証して unrecognized_model で失敗する。
Qwen models, 2.6-3.5% slower decode for DeepSeek-V4 and GLM-5.3, and 1-3%
AI-generated: developed with the DeepSeek Harness agent (model GLM-5.3 Flash); human-directed — SySL-1.0 §2.3 disclosure.
Model: GLM-5.3 (zai-individual-coding-plan)
Model: GLM-5.3 (account:zai-individual-coding-plan)
Model: account:zai-individual-coding-plan/GLM-5.3 Operator-request: step 3 of #451 — split src/tg/mod.ts with pure moves
Model: GLM-5.3 (zai-individual-coding-plan)
refused twice (release validation C8/C9 on GLM-5.3). The refusal now quotes
The release validation ran /dev-plan on GLM-5.3 (Z.AI): at MaxTokens 24576
GLM-5.3: the text model of GLM-5.3-Flash (glm5_next) and GLM-5.3 (glm_moe_dsa), with a tiny-checkpoint reference check.
GLM-5.3: the chat template (reasoning effort line, thinking, tool calls) and the command line, engine and streaming wiring.
GLM-5.3: catalog entries, docs/glm.md with the measured agreement (transformers on tiny checkpoints, llama.cpp on the real UD-Q2_K_XL), the models table and the README index; glm5_check
(Max)", "Muse Spark 1.3 (Max)", "Gemini 3 Pro", "GLM-5.3 (Max)", "Qwen3.7-Max-Preview", "Kimi K3 (Max)", "Muse Spark", "Qwen3.5-Max-Preview", … (+159 names)
even on a cache hit (measured on z.ai glm-5.3; one decide call completed upstream in 25.8s and was billed, while bili had already aborted at
- GLM by Z.ai, model id `z-ai/glm-5.3`, run through opencode by the Paperclip Runner. Extended reasoning enabled, with shell and file tool use.
z.ai GLM (Coding Plan), model id `zai/glm-5.3` (approx. 200K context), agentic tool-use mode — used for source analysis, the fix, and the test.
- промпт в `.scratch/prompt.md`, запуск `opencode run --model opencode/glm-5.3 --auto`;
SDK scans. Users can select `--provider huggingface --model zai-org/GLM-5.3:novita` with `HF_TOKEN` instead of OpenAI authentication.
conversation, tools, references, prompts, model `glm-5.3`, token budgets, labels
claude-code 71, codex-cli 46, antigravity 6, openai-oauth 67, glm-5.3-flash 12, glm-5.3 12.
<li>28e30cd: support low and max reasoning effort for GLM-5.3 models</li>
model: ollama/glm-5.3:cloud
Qwen3.8-2.4T, Qwen3.8-27B, Gemma 4 31B, DeepSeek-V4.1-Flash, DeepSeek-V4-Pro, Kimi-K3.1-Pro, GLM-5.3, MiniMax-M3.1-Flash, Mistral-Large-4, Muse-Spark-Open, Nemotron-3.5-Super, Hy4-preview; small: LFM3-1.7B, Ling-3.0-tiny, granite-4.2-3b, granite-4.2-8b, Qwen3.5-9B, LFM2.5-24B, Nemotron-3.5-Lightning, Gemma 4 26B,
Implementer: agent=opencode model=zai-coding-plan/glm-5.3 effort=high
- Model: enmaas/rits/zai-org/glm-5-3@high
- Model: enmaas/rits/zai-org/glm-5-3@high
a dependency-and-fence graph, drops Windows support, and adds GLM-5.3/Kimi K3 LLM support with fused GPT-OSS kernels.
Follow-up worker models: zai/glm-5.3 (high thinking), lane glm
Switch High to `z-ai/glm-5.3`, replace the slow Medium fallback and an unreachable worker fallback, refresh every catalog price from the live OpenRouter US models API,
chat models of that catalog on 2026-11-10: `deepseek-ai/DeepSeek-V4.1-Flash`, `zai-org/GLM-5.3-Flash`, `deepseek-ai/DeepSeek-V4-Pro-0813` and `zai-org/GLM-5.3`. The first is small and generally available, as the default has to be.
- Models: `zai-coding-plan/glm-5.3` (`low`)
`type: "message"` the Provider API adapter requires for `zai-org/GLM-5.3`, which answered `400 invalid_request: input item is missing type` on the first turn. The rewrite is
#36 has two cross-model reviews GREEN (Claude and zai/glm-5.3).
(new): Pick a model (keeps the `qualified-models` anchor), GLM-5.3 one machine or two, Models on disk, Model discovery. Table rows are copied verbatim: only relative
approved Unlid backup. Provider entries retain the existing `zhipuai/glm-5.3` and `zhipuai/glm-5.3-flash` lineage while explicitly identifying the served variants.
| GLM-5.3 (`GlmMoeDsaForCausalLM`, same class), fp8 KV | Yes, but its default RoPE accepts `key=None`, so it never crashed | GSM8K, 1319 questions | 0.901
Chatwoot 2171 (2026-11-05) — eight non-streaming `private/kimi-k3` / `private/glm-5-3` answers of 9–22 minutes, generated and billed in full ($10.76), delivered to nobody; the same caller's
During the release validation (C10), `/dev-plan` on GLM-5.3 refused the same plan twice with:
During the release validation (C8, C9), `/dev-plan` on GLM-5.3 refused the same plan twice with
The release validation (C8) ran `/dev-plan` on GLM-5.3 through Z.AI:
vision tool — e.g. a text-only main model (GLM-5.3) pinned to a vision-capable describer (GLM-5.3-Flash). But `decide_image_input_mode` applies that pin globally: a delegated child spawned
- the text model of GLM-5.3-Flash (glm5_next) and GLM-5.3 (glm_moe_dsa), with a tiny-checkpoint reference check.
Four nodes, GLM-5.3 Mixed346 with its runtime adapter, production's recipe. Twelve prompts of 4K to 12K tokens, sent one at a time, each carrying a
| Chutes | `api.chutes.ai/v1` (404) | `llm.chutes.ai/v1`, `zai-org/GLM-5.3` | reaches the API; 402 = $0 balance |
https://github.com/BerriAI/litellm/pull/51231. It covers SDK and gateway setup for GLM-5.3 and GLM-5.3 Flash, streaming usage, tool-result continuations, model limits, cached-input pricing and timeout handling.
— hive: agent=quality backend=curvebender model=rits/zai-org/glm-5-3
— hive: agent=quality backend=curvebender model=rits/zai-org/glm-5-3
arsePresetsForTest( Object.fromEntries( plans.map(([id, , api]) => [ id, { name: 'Ambiguous upstream label', npm: '@ai-sdk/openai-compatible', api, models: { 'glm-5.3': { name: 'GLM-5.3', tool_call: true } }, }, ]), ), ); expect(presets).toHaveLength(4); for (const [providerId, name, baseUrl] of plans) { const preset = presets.find((item) => item.providerId === providerId); expect(preset).toMatchObject({ providerId, name, baseUrl, apiFormat: 'openai-completions', models: [{ modelId: 'glm-5.3' }], }); expect(providerCompletionUrl('openai-completions', preset!.baseUrl)).toBe( `${baseUrl}/chat/completions`, ); } }); it('builds each Provider icon URL from the catalog snapshot pr
Это упоминание в коде не подтверждает рабочий идентификатор модели.
GLM-5.3 — модель Z.ai с открытыми весами, выпущенная 18 авг. 2026. Контекстное окно — 1 048 576 токенов. Максимальная длина ответа — 131 072 токена. Цена за миллион токенов: $1,40 на входе и $4,40 на выходе.
Контекстное окно — 1 048 576 токенов. Максимальная длина ответа — 131 072 токена.
Цена за миллион токенов: $1,40 на входе и $4,40 на выходе. Кэшированный вход — $0,26 за миллион токенов.
Да. Z.ai публикует веса модели.
Дата релиза — 18 авг. 2026.