LLM Comparison
GLM-5.3-Flash vs Step 3.5 Flash
Side-by-side specs, pricing & capabilities
Add to comparison
2/6 models| Organization | ||
| OpenTools Score | 74 456 | |
| Family | GLM | Step |
| Status | Current | Current |
| Release Date | Aug 2026 | Jan 2026 |
| Context Window | 1.3M tokens | 262K tokens |
| Input Price | $0.07/M tokens | $0.10/M tokens |
| Output Price | $0.25/M tokens | $0.30/M tokens |
| Pricing Notes | OpenRouter lists a limited-time discounted price of $0.075/M input and $0.25/M output. Standard first-party style pricing is commonly $0.15/M input and $0.50/M output; cache-read pricing may vary by provider. | — |
| Capabilities | textvisionvideo-inputcodetool-usereasoninglong-contextopen-weights | textcode |
| Training Cutoff | Not disclosed | — |
| Max Output | 131K tokens | 66K tokens |
| API Identifier | z-ai/glm-5.3-flash | stepfun/step-3.5-flash |
| Benchmarks | ||
| Artificial Analysis Intelligence Index v4.3 | 41.9 indexartificial-analysis | — |
| GPQA Diamond | 91.2%artificial-analysis | — |
| SciCode | 51.6%artificial-analysis | — |
| Terminal-Bench 2.1 | 84.3%artificial-analysis | — |
| View GLM-5.3-Flash | View Step 3.5 Flash | |
Cost Calculator
Enter your expected monthly token usage to compare costs.
| Model | Input | Output | Total / mo | vs Best |
|---|---|---|---|---|
| GLM-5.3-FlashCheapest | $0.08 | $0.13 | $0.20 | — |
| Step 3.5 Flash | $0.10 | $0.15 | $0.25 | +25% |
Compare hosting configurations
Compare two exact setups, including different configurations of the same model. Hardware fit, measured speed and costs are separate decisions.
Add your cost assumptions (optional)
Use hours for the same scenario period. Cloud billing includes idle time; active hours alone do not set the bill. Local electricity uses whole-system attributable power, powered hours and your tariff. Missing inputs remain unknown.
Measured speed needs the same pinned workload protocol, tokenizer, context, batching and cache state.
| Configuration | Run full native GLM-5.3-Flash image, video and reasoning on an owned DGX H200 TP4 subset | Host full native Step 3.5 Flash reasoning and tools on an owned eight-H100 server |
|---|---|---|
| Model | GLM-5.3-Flash | Step 3.5 Flash |
| Hardware | Owned NVIDIA DGX H200; 4 of 8 GPUs, diagnostic TP4 | Existing eight-H100 80 GB GPU server |
| Runtime | GLM 5.3 Flash full native NVIDIA immutable image Official image 0674a3ea, declared vLLM 385dce36 / Python 3.12 / CUDA 13.0.2; reader source guards required · CUDA | Released native Step3.5 Flash BF16 with three-layer MTP on vLLM vLLM 0.27.1 commit 6e448d0; exact official Linux amd64 image digest, reader preflight required · CUDA |
| Artifact | GLM-5.3-Flash complete native mixed FP8/BF16 multimodal checkpoint · Native mixed FP8 with the exact BF16 vision, attention, KDA and mHC exclusions retained · Native FP8 language experts with exact BF16 exclusions and full BF16 vision; no re-quantization/component drop | Full native Step 3.5 Flash BF16 with three MTP layers · Complete native BF16 weights, including every expert and MTP layers 45,46,47 · Unquantized native BF16; complete experts and all MTP tensors retained |
| Context / concurrency | 16,384 tokens / 1 | 8,192 tokens / 1 |
| Required memory | Unknown (unknown) | Unknown (unknown) |
| Evidence | Documented setup | Documented setup |
| Hardware fit | Hardware fit unknown Your operating system is unknown. Runtime support has not been confirmed. Full weights, BF16 vision, KDA recurrent/convolution states, sparse MLA, mHC, MTP, collectives, kernels and caches have not been measured. Enter available memory per device or unified memory pool. Host RAM: Preserve the DGX vendor 2 TB and AWS 2,048 GiB source labels. Reader records actual MemTotal; no nominal-to-usable threshold or measured minimum is asserted. Free disk space: A 1 TiB usable working-space guard is prescribed for assets, image/staging and results. It is a conservative reader allocation, not a measured minimum or proof the unpacked runtime fits. Container requirement: Separate immutable Linux amd64 container; host/container versions not conflated | Hardware fit unknown No measured per-device or whole-process peak for full BF16/MTP3 on eight H100s at8192 context. Enter available memory per device or unified memory pool. Host RAM: 2048 GiB host allocation is a planning choice, not measured minimum. Free disk space: 768 GiB allocated disk and600 GiB free-space guard are explicit prescriptions; complete installed runtime/cache requirements remain unmeasured. |
| One-time cost | Unknown Existing hardware/allocation: Existing owned acquisition cost is unknown. Reader setup labor: Reader download, image pull, setup and labor costs are unknown and separate from recurring use. | Unknown hardware: Existing owned GPU/host purchase and incremental cost are unknown. setup: Reader setup and labor cost is unknown. |
| Recurring scenario cost | Unknown Owned electricity: Owned-host electricity/power price and measured use are unknown. Retained local/storage: Existing NVMe retention/opportunity cost is unknown. Instance-store data may be lost on stop or termination; copy wanted results first. Network/egress: Network, data-transfer charges and taxes are unknown; no free assumption. Administration: Reader-specific labor and administration costs are unknown. One-time setup and recurring costs are separate. Unknown components are not treated as free. No monthly token output is inferred from peak throughput. | Unknown electricity: Actual wall power/rate/poweredhours unknown; no GPU TDP conversion. storage: Incremental owned 768 GiB storage costs are unknown. egress: Actual transfer/allowance/traffic cost unknown. administration: Maintenance time/rate unknown. One-time setup and recurring costs are separate. Unknown components are not treated as free. No monthly token output is inferred from peak throughput. |
| Published measurements (unranked) | No compatible measurement available | No compatible measurement available |
Different currencies and incomplete costs are not ranked. Model quality benchmarks remain on the model pages; they do not establish quality or speed for every deployment.
Z.ai
GLM-5.3-Flash
GLM-5.3-Flash is Z.ai’s open-weight GLM-5 model for efficient coding, long-horizon agent tasks, and long-context multimodal reasoning. It uses a hybrid sparse and linear attention design with 320B total parameters and 18B active parameters.
StepFun
Step 3.5 Flash
Step 3.5 Flash is a large language model from StepFun. Supports up to 262,144 token context window. Available from $0.10/M input tokens.
More Comparisons
Looking for more AI models?
Browse All LLMs