GLM 5.3 Flash
z-ai/glm-5.3-flashHugging FaceChatVisioncybersecurityreasoningtoolsmoeGLM-5.3-Flash — 195 GB on disk (down from the 643 GB BF16 source) — INT4 W4A16 quantization of the routed experts in compressed-tensors pack-quantized format, for Hopper (H100 / H200) via vLLM Marlin kernels.
Versions
Every served version of this family. Each has its own providers and prices.
- GLM 5.3 Flashz-ai/glm-5.3-flashbase$0.031 in$0.124 out1M context1 endpoint
- GLM 5.3 Flash Cybersecurity (DealignAI)z-ai/glm-5.3-flash-cybersecurity$0.0465 in$0.155 out1M context2 endpoints
DealignAI's cybersecurity build of GLM 5.3 Flash, with refusals removed, in W4A16 (int4 experts, BF16 activations). It takes text, image and video input, reasons, calls tools and returns structured JSON. Each request has a 1,048,576-token context limit with at most 65,536 output tokens. Its template knows three reasoning levels: minimal and low run at Low, medium and high at High, xhigh at Max; thinking cannot be switched off. Its safety guardrails are substantially reduced.
Providers
Live endpoints, cheapest first. First token is the median wait for a streamed reply; throughput covers the last day and uptime the last 24 hours.
- Consensus ProtocolHealthyint469% offUSbase
- Input
- $0.031
- Output
- $0.124
- Context
- 1M66K out
- First token
- 347mslast 30 min
- Throughput
- 39 tok/s
- Uptime 24h
- 100.0%
- ImpCCHealthyint469% offUScybersecurity
- Input
- $0.0465
- Output
- $0.155
- Context
- 1M66K out
- First token
- 56.7slast 30 min
- Throughput
- 60 tok/s
- Uptime 24h
- 100.0%
- Consensus ProtocolHealthyint469% offUScybersecurity
- Input
- $0.124
- Output
- $0.496
- Context
- 1M66K out
- First token
- 648mslast 30 min
- Throughput
- 39 tok/s
- Uptime 24h
- 99.8%
| Provider | Context | Input | Output | First token | Throughput | Uptime 24h |
|---|---|---|---|---|---|---|
| Consensus ProtocolHealthyint469% offUnited Statesbase | 1M66K max out | $0.031 | $0.124 | 347mslast 30 min | 39 tok/s | 100.0% |
| ImpCCHealthyint469% offUnited Statescybersecurity | 1M66K max out | $0.0465 | $0.155 | 56.7slast 30 min | 60 tok/s | 100.0% |
| Consensus ProtocolHealthyint469% offUnited Statescybersecurity | 1M66K max out | $0.124 | $0.496 | 648mslast 30 min | 39 tok/s | 99.8% |
Capabilities and data
What each endpoint declares about tools, reasoning and prompt data, beside the marketplace’s own checks. The key below says what each badge means.
- Consensus Protocolint4baseTools VerifiedReasoning Effort levelsData Zero retention
- ImpCCint4cybersecurityTools VerifiedReasoning Effort levelsData Zero retention
- Consensus Protocolint4cybersecurityTools VerifiedReasoning Effort levelsData Zero retention
- Verified
- a current check called an offered tool with valid arguments
- Effort levels
- effort levels are declared, and none is failing checks
- Zero retention
- prompts are neither stored nor trained on; the claim passed marketplace review
Pricing
USD per million tokens.
| Provider | Input | Output | Cache read | Cache hit | Effective input |
|---|---|---|---|---|---|
| Consensus Protocolint469% offbase | $0.031 | $0.124 | $0.0155 | 82.7% | $0.018177 |
| ImpCCint469% offcybersecurity | $0.0465 | $0.155 | $0.0093 | 92.2% | $0.012209 |
| Consensus Protocolint469% offcybersecurity | $0.124 | $0.496 | $0.031 | 84.7% | $0.045201 |
Price history
Input price · USD per million tokens
Output price · USD per million tokens
Performance
Timing percentiles cover the labelled window: the last 30 minutes when it has enough samples, otherwise the last day. TTFT is the seconds to the first token, measured only on streamed requests; response is the seconds from request to last byte, measured on every request, streamed or not. Throughput is output tokens per second over the last day, counted only over streamed generations long enough to have a rate — a buffered answer has no generation phase, and a two-token one times the scheduler rather than the model. Tool calls is the share of the last day of tool requests this endpoint answered with a tool call, published once the day holds twenty of them; under that the cell is blank rather than a rate nothing measured.
| Provider | TTFT p50 | p75 | p90 | p99 | Response p50 | p75 | p90 | p99 | Throughput p50 | p75 | p90 | p99 | Tool calls 1d | Runaways 1d | Empty 1d | Capacity |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Consensus Protocolint4base | 347mslast 30 min | 697ms | 726ms | 733ms | 1.51slast 30 min | 20.5s | 2m 49s | 5m 41s | 39 tok/s | 57 tok/s | 82 tok/s | 158 tok/s | — | 0.8% | 0.0% | 34/35 free |
| ImpCCint4cybersecurity | 56.7slast 30 min | 1m 3s | 1m 6s | 1m 37s | 10.0slast 30 min | 49.7s | 1m 38s | 5m 51s | 60 tok/s | 67 tok/s | 74 tok/s | 89 tok/s | — | 0.0% | 0.1% | 13/16 free |
| Consensus Protocolint4cybersecurity | 648mslast 30 min | 1.32s | 4.43s | 4.43s | 883mslast 30 min | 2.20s | 4.52s | 6.73s | 39 tok/s | 44 tok/s | 49 tok/s | 58 tok/s | — | 0.0% | 0.0% | 8/8 free |
Time to first token by hour · last 7 days
Hourly median per provider. The line spans hours with no streamed traffic; where samples are sparse, a dot marks each hour actually timed.
Throughput by hour · last 7 days
Hourly median output tokens per second per provider, over generations long enough to have a rate. The line spans hours with no answer long enough to measure; where samples are sparse, a dot marks each hour actually measured.
Benchmarks
Scores from the benchmark board, one column per served version. Each figure comes from the endpoint with the most evidence for that suite, and a provisional one names how much of its suite it covers.
Uptime
Share of successful requests and probes over the last day and week. Reliability is the share of the requests a provider accepted that it answered; turned away is the share it refused with its own 429 because it was full, which uptime counts as failures.
- Consensus Protocolint4base1d 100.0%7d 99.9%reliability 1d 100.0%7d 99.9%turned away 1d 0.0%7d 0.1%Healthy
- ImpCCint4cybersecurity1d 100.0%7d 99.9%reliability 1d 100.0%7d 99.9%turned away 1d 0.0%7d 0.0%Healthy
- Consensus Protocolint4cybersecurity1d 99.8%7d 99.9%reliability 1d 99.9%7d 99.9%turned away 1d 0.1%7d 0.1%Healthy
Availability by hour · last 7 days
Share of successful requests and probes per hour and provider. The line spans hours with nothing to measure; where samples are sparse, a dot marks each hour actually measured.
Activity
Prompt tokens and completion tokens per UTC day over the last 30 days. Every billed request counts, the platform's own paid probe traffic included.
curl https://api-staging.heimdallm.com/api/v1/chat/completions \ -H "Authorization: Bearer $HEIMDALL_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "z-ai/glm-5.3-flash", "messages": [{ "role": "user", "content": "Hello" }] }'Set HEIMDALL_API_KEY to a key from your account. See the quickstart for routing suffixes and provider preferences.
About
- Author
- Z.ai
- Family id
- z-ai/glm-5.3-flash
- Weights
- zai-org/GLM-5.3-Flash
Z.ai: GLM 5.3 FlashReleased Aug 30, 2026
GLM-5.3-Flash — 195 GB on disk (down from the 643 GB BF16 source) — INT4 W4A16 quantization of the routed experts in compressed-tensors pack-quantized format, for Hopper (H100 / H200) via vLLM Marlin kernels.
- Weights
- JANGQ-AI/GLM-5.3-Flash-W4A16
- Request parameters
- frequency_penalty (-2–2), max_tokens (1–65536), min_p (0–1), presence_penalty (-2–2), reasoning_effort (none, low, medium, high), repetition_penalty (0–2), response_format (any value), seed (0–2147483647), stop (list), temperature (0–2), tool_choice (any value), tools (true/false), top_k (0–200), top_p (0–1)
- Licence
- glm-5.3
- Size
- 321.3B parameters, I32 weights
- Tokenizer
- —
- Available until
- —
Z.ai: GLM 5.3 Flash Cybersecurity (DealignAI)Released Aug 30, 2026
DealignAI's cybersecurity build of GLM 5.3 Flash, with refusals removed, in W4A16 (int4 experts, BF16 activations). It takes text, image and video input, reasons, calls tools and returns structured JSON. Each request has a 1,048,576-token context limit with at most 65,536 output tokens. Its template knows three reasoning levels: minimal and low run at Low, medium and high at High, xhigh at Max; thinking cannot be switched off. Its safety guardrails are substantially reduced.
- Request parameters
- add_generation_prompt (true/false), allowed_token_ids (list), cache_salt (text), chat_template_kwargs (object), continue_final_message (true/false), frequency_penalty (-2–2), guided_choice (list), guided_grammar (text), guided_json (any value), guided_regex (text), ignore_eos (true/false), include_stop_str_in_output (true/false), logprobs (true/false), max_completion_tokens (1–65536), max_tokens (1–65536), min_p (0–1), min_tokens (0–65536), n (1–1), parallel_tool_calls (true/false), presence_penalty (-2–2), reasoning (object), reasoning_effort (varies by provider), repetition_penalty (varies by provider), response_format (varies by provider), return_tokens_as_token_ids (true/false; requires logprobs=true), seed (varies by provider), stop (varies by provider), stop_token_ids (list), structured_outputs (true/false), temperature (0–2), tool_call_dispatch (true/false), tool_choice (varies by provider), tools (varies by provider), top_k (varies by provider), top_logprobs (0–20; requires logprobs=true), top_p (varies by provider)
- Licence
- mit
- Size
- 321.3B parameters, I32 weights
- Tokenizer
- —
- Available until
- —