> For clean Markdown of any page, append `.md` to the page URL. > For a complete documentation index, see https://docs.sarvam.ai/llms.txt. > For full documentation content in one file, see https://docs.sarvam.ai/llms-full.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.sarvam.ai/_mcp/server. # GLM-5.2 > GLM-5.2 is an open-source model served on Sarvam infrastructure. 512K token context window with tool calling and visible reasoning. GLM-5.2 is an [open-source model](/api/getting-started/models/open-source) served by Sarvam in **beta**, and is **not tuned for Indian languages**. Beta access is granted per API key — [contact us](/api/getting-started/help) to request whitelisting. For Indian language workloads use [Sarvam-105B](/api/getting-started/models/sarvam-105b) instead. A general-purpose open-source model. Its distinguishing feature on Sarvam is a very large **512K token context window** — four times the 128K offered by Sarvam's own chat models — which makes it a reasonable choice for reasoning over long documents or large codebases in a single request. ## At a Glance | | | | ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Model ID** | `glm5.2` | | **What it does** | General-purpose chat LLM; very long context and tool calling | | **Languages** | Not tuned for Indian languages by Sarvam | | **APIs** | [Open-Source Models API Reference](/api-reference/open-source/chat-completions) (`POST /v2/chat/completions`, OpenAI-compatible, streaming supported) — also see [using an open-source model](/api/getting-started/models/open-source#using-an-open-source-model) | | **Input limits** | 524,288-token context window; 10 MB request cap — [all limits](#limits-and-errors) | | **Benchmarks** | None published by Sarvam | | **Pricing** | ₹128.1 input · ₹23.79 cached input · ₹402.6 output — per 1M tokens; reasoning billed as output ([Pricing](/api/getting-started/pricing)) | | **Best for** | Long-document and large-codebase analysis in a single request | | **Not for** | Indian language workloads — use [Sarvam-105B](/api/getting-started/models/sarvam-105b) | ## Why you might use it #### Very long context A 524,288 token context window lets you pass entire books, long transcripts, or large code repositories in one request without chunking or retrieval. #### Tool calling Supports the OpenAI-compatible `tools` and `tool_choice` parameters, so it works with agentic loops and function-calling frameworks. ## Model Specifications #### Key Considerations ## GLM-5.2 vs Sarvam's chat model | Feature | GLM-5.2 | Sarvam-105B | | -------------------------- | ----------------------- | --------------------------------------------- | | **Context window** | 524,288 | 128,000 | | **Tool calling** | ✅ | ✅ | | **Image input** | ❌ | ❌ | | **Indian language tuning** | None | 23 languages | | **Best for** | Very long-context tasks | Maximum quality reasoning & agentic workflows | Pick GLM-5.2 **only** when your task genuinely needs more than 128K of context. For everything else — and especially for anything touching Indian languages — a Sarvam model will serve you better and comes with Sarvam's own evaluation behind it. ## Key Capabilities #### Basic Chat Completion A single-turn request. Only the `model` field differs from a Sarvam chat completion call. #### Python ```python from sarvamai import SarvamAI client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY") response = client.open_source_models.chat_completions_v2( model="glm5.2", messages=[ {"role": "user", "content": "Explain the difference between a B-tree and a B+ tree index."} ], temperature=0.2, top_p=1, max_tokens=2000, ) print(response.choices[0].message.content) ``` #### JavaScript ```javascript import { SarvamAIClient } from "sarvamai"; const client = new SarvamAIClient({ apiSubscriptionKey: "YOUR_SARVAM_API_KEY", }); async function main() { const response = await client.openSourceModels.chatCompletionsV2({ model: "glm5.2", messages: [ { role: "user", content: "Explain the difference between a B-tree and a B+ tree index.", }, ], temperature: 0.2, top_p: 1, max_tokens: 2000, }); console.log(response.choices[0].message.content); } main(); ``` #### cURL ```bash curl -X POST https://api.sarvam.ai/v2/chat/completions \ -H "api-subscription-key: $SARVAM_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "messages": [ {"role": "user", "content": "Explain the difference between a B-tree and a B+ tree index."} ], "model": "glm5.2", "temperature": 0.2, "top_p": 1, "max_tokens": 2000 }' ``` The answer is in `message.content`. The model's chain-of-thought arrives separately in `message.reasoning_content` — see [Reasoning](#reasoning) below for how to budget for it or turn it off. #### Long-Context Analysis GLM-5.2's main advantage. Pass a large document directly in the prompt instead of chunking it — up to 512K tokens of context, minus whatever you reserve for `max_tokens`. #### Python ```python from sarvamai import SarvamAI client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY") with open("annual_report.txt", encoding="utf-8") as f: document = f.read() response = client.open_source_models.chat_completions_v2( model="glm5.2", messages=[ {"role": "system", "content": "You are a financial analyst. Answer only from the document provided."}, {"role": "user", "content": f"{document}\n\nList every material risk disclosed, with the page or section it appears in."}, ], temperature=0.1, max_tokens=4000, ) print(response.choices[0].message.content) ``` #### cURL ```bash # Build the payload with a file so large documents do not hit shell argument limits. python3 - <<'PY' > payload.json import json document = open("annual_report.txt", encoding="utf-8").read() json.dump({ "model": "glm5.2", "messages": [ {"role": "system", "content": "You are a financial analyst. Answer only from the document provided."}, {"role": "user", "content": document + "\n\nList every material risk disclosed."}, ], "temperature": 0.1, "max_tokens": 4000, }, open("payload.json", "w")) PY curl -X POST https://api.sarvam.ai/v2/chat/completions \ -H "api-subscription-key: $SARVAM_API_KEY" \ -H "Content-Type: application/json" \ --data-binary @payload.json ``` Requests are capped at **10 MB** regardless of the context window. A document that fits in 512K tokens can still exceed the byte cap and return `413 Payload Too Large`. #### Tool Calling GLM-5.2 supports OpenAI-compatible function calling. #### Python ```python from sarvamai import SarvamAI client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY") response = client.open_source_models.chat_completions_v2( model="glm5.2", messages=[ {"role": "user", "content": "What is the weather in Bengaluru right now?"} ], tools=[ { "type": "function", "function": { "name": "get_weather", "description": "Get the current weather for a city.", "parameters": { "type": "object", "properties": { "city": {"type": "string", "description": "City name"} }, "required": ["city"], }, }, } ], tool_choice="auto", max_tokens=1000, ) print(response.choices[0].message.tool_calls) ``` #### cURL ```bash curl -X POST https://api.sarvam.ai/v2/chat/completions \ -H "api-subscription-key: $SARVAM_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "glm5.2", "messages": [ {"role": "user", "content": "What is the weather in Bengaluru right now?"} ], "tools": [ { "type": "function", "function": { "name": "get_weather", "description": "Get the current weather for a city.", "parameters": { "type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"] } } } ], "tool_choice": "auto", "max_tokens": 1000 }' ``` #### Streaming Set `stream` to `true` to receive tokens as server-sent events. Because GLM-5.2 reasons before answering, the stream delivers `delta.reasoning_content` chunks first and `delta.content` chunks after — read both, or the wait before the first `content` chunk looks like a stall. #### Python ```python from sarvamai import SarvamAI client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY") stream = client.open_source_models.chat_completions_v2( model="glm5.2", messages=[ {"role": "user", "content": "Write a short note on database normalisation."} ], max_tokens=1000, stream=True, ) for chunk in stream: delta = chunk.choices[0].delta reasoning = getattr(delta, "reasoning_content", None) if reasoning: pass # the chain-of-thought — show it, log it, or skip it if delta.content: print(delta.content, end="", flush=True) ``` #### cURL ```bash curl -N -X POST https://api.sarvam.ai/v2/chat/completions \ -H "api-subscription-key: $SARVAM_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "glm5.2", "messages": [ {"role": "user", "content": "Write a short note on database normalisation."} ], "max_tokens": 1000, "stream": true }' ``` The `usage` object reported on **streamed** GLM-5.2 responses is currently unreliable — it can report far fewer completion tokens than were actually generated and billed. For accurate token accounting, use a non-streaming request or the usage page on the [dashboard](https://dashboard.sarvam.ai). ## Reasoning GLM-5.2 thinks before every answer — **by default**, not only when `reasoning_effort` is set. The chain-of-thought arrives in a separate `reasoning_content` field on the message (`delta.reasoning_content` when streaming); `content` carries only the final answer. Two practical consequences: * **Reasoning tokens are billed and count against `max_tokens`.** A small budget can be consumed entirely by reasoning, returning `content: null` with `finish_reason: "length"`. If your answers come back empty, raise `max_tokens` or turn thinking off. * **Turn thinking off when you set `response_format`.** With reasoning active, a JSON-constrained request returns `content: null` instead of JSON. The off-switch goes through `extra_body`: ```json { "model": "glm5.2", "messages": [{ "role": "user", "content": "..." }], "extra_body": { "chat_template_kwargs": { "enable_thinking": false } } } ``` With thinking off, responses are faster, the whole `max_tokens` budget goes to the answer, and Structured Outputs work reliably. ## Parameters GLM-5.2 accepts the standard OpenAI-compatible parameter set — `temperature`, `top_p`, `max_tokens`, `stream`, `stop`, `n`, `seed`, `frequency_penalty`, `presence_penalty`, `reasoning_effort`, `response_format` (Structured Outputs), `tools`, `tool_choice`, and `extra_body`. See the [full parameter table](/api/getting-started/models/open-source#supported-parameters) for types, defaults, and ranges. Model-specific exceptions: * **`image_url` content parts are rejected** (`400`) — GLM-5.2 is text only. * **`response_format` requires thinking off** — see [Reasoning](#reasoning) above. * **`seed` is not currently reproducible** on this model — identical seeds return different completions. Use `temperature: 0` where you need stability. ## Limits and errors Common errors for every model on `/v2/chat/completions` are documented on the [Open-Source Models overview](/api/getting-started/models/open-source#errors). GLM-5.2-specific cases: | Condition | Status | Code | What it means | | ---------------------------------------------------------------------- | ------ | ---------------------------- | --------------------------------------------------------------------------------------------------------------- | | API key lacks beta access | `400` | `invalid_request_error` | Request whitelisting — see [Beta APIs](/api-reference/beta-apis). | | Missing or invalid API key | `403` | `invalid_api_key_error` | Check the `api-subscription-key` header. | | No credits remaining | `402` | `insufficient_quota_error` | Top up on the [dashboard](https://dashboard.sarvam.ai). | | Unknown or unavailable model id | `404` | `not_found_error` | Confirm the id with [`GET /v2/models`](/api-reference/open-source/models). | | Message contains an image part | `400` | `invalid_request_error` | GLM-5.2 is text only. Use [`gemma4`](/api/getting-started/models/open-source/gemma-4-31b) for image input. | | `max_tokens` above 524,288 | `400` | `invalid_request_error` | `max_tokens` cannot exceed the context window. | | Estimated prompt tokens + `max_tokens` above 524,288 | `422` | `unprocessable_entity_error` | Shorten the prompt or lower `max_tokens`. | | Request body above the model's size cap | `413` | `invalid_request_error` | The byte cap is independent of the token window. | | `tool_choice` naming a function with no `tools` | `422` | `unprocessable_entity_error` | Provide the `tools` array alongside `tool_choice`. | | Invalid tool schema (`$ref`, recursion, size) | `400` | `invalid_request_error` | See [Tool schemas](/api/getting-started/models/open-source#tool-schemas). | | Rate limit exceeded | `429` | `rate_limit_exceeded_error` | Back off and retry; see [Rate Limits](/api/getting-started/ratelimits). | | Model overloaded after retries | `503` | `model_overloaded` | Retry later. | | `content: null` with `finish_reason: "length"` | `200` | — | Reasoning consumed the whole `max_tokens` budget — raise it, or turn thinking off. See [Reasoning](#reasoning). | | `content: null` with `finish_reason: "stop"` and `response_format` set | `200` | — | Thinking + Structured Outputs conflict — turn thinking off. See [Reasoning](#reasoning). | ## Support Sarvam supports the serving layer — auth, billing, rate limits, availability. Sarvam does not tune or evaluate this model, so its output quality, reasoning, and language coverage are properties of the model itself. See [Open-Source Models](/api/getting-started/models/open-source) for the full support boundary, and [Talk to us](/api/getting-started/help) for serving issues.