Evaluate your AI stack on real code
Compare models, harnesses, and tools on consistent repository tasks to understand quality, cost, latency, and behavior before making infrastructure decisions.
Learn more about mo evalMomento Inference
Choose models that perform on your workflows. Momento serves them from the region you pick, on a stack built for speed and optimized for every model.
AMD and NVIDIA GPUS Available now
The tuned inference stack
Why Momento Inference
Model choice is only the beginning. Bring your workload, and optimize the stack around it.
Code check-ins, file moves, and ticket triage can each run on a different model. Each chosen by how it performs on your own tasks. The diagram shows example assignments.
See the benchmarksMomento optimizes inference for your workload to reduce token usage and lower the cost per task. Built on our expertise in KV caching, Momento also retrieves repeated context faster, with cached input costing up to 90% less. The chart shows a hypothetical token reduction.
Work with our inference teamStandard workload
Momento-optimized workload
Dedicated endpoints serve only you, and shared endpoints keep each customer's data secret. Pick the region, and the gateway enforces zero data retention on every call.
Ask about dedicated capacityWorks with LiteLLM and the mo gateway.
Models & pricing
Featured models. Transparent prices per 1M tokens.
No sales call needed. Sign up, add a card, and start calling any model in the catalog.
Per 1M tokens. Cached input applies when a request reuses context.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GLM-5.3 Flash NEW | $0.135 | $0.027 | $0.45 |
| GLM-5.3 Flash Fast NEW* | $0.54 | $0.108 | $1.80 |
| GLM-5.3 | $1.40 | $0.14 | $4.40 |
| DeepSeek V4 Pro | $1.32 | $0.132 | $3.96 |
| Gemma 4 Tailwind | $0.12 | $0.12 | $0.35 |
Kimi K3 and additional Claude models are also available. Explore the catalog for all models and cache-write pricing.
* GLM-5.3 Flash Fast uses 4-bit MXFP4 quantization with a 1M-token context window. When Momento capacity is full, requests overflow to partner inference providers at the same price. Approximately 735 tokens per second in a single session, measured November 17, 2026.
Beyond inference
Momento goes beyond serving models with tools to evaluate your AI stack against real workloads and manage how models are used in production.
Compare models, harnesses, and tools on consistent repository tasks to understand quality, cost, latency, and behavior before making infrastructure decisions.
Learn more about mo evalManage budgets, approved-model routing, and security policies across inference requests, or use Momento Inference behind your existing LiteLLM router.
Explore mo gatewayBuilt by the team behind infrastructure
trusted in production
Your next workload starts here
Start building with Momento Inference on your current models running on our hardware. Bring our engineers in when you need to go further.