Tensor Machine Docs

Models Reference

How models work on the Tensor Machine API — a single OpenAI-compatible endpoint, India-resident. The live catalogue and current pricing are on the website.

Available models

Tensor Machine serves a catalogue of frontier models via a single OpenAI-compatible API endpoint. All models run on India-resident infrastructure.

The examples below are illustrative — the live catalogue is larger and changes often, and current pricing lives on the website, not here. For the up-to-date model list, context windows, and per-model INR rates, see tensormachine.ai/models and tensormachine.ai/pricing.

Example models

A few representative models and the kinds of work they suit (not the full list, and not pricing — check the live catalogue for both):

Model IDProviderContextType
z-ai/glm-5.2Z-AI1MChat, Code
minimax/minimax-m3MiniMax1MChat, Code, Vision
deepseek/deepseek-v4-proDeepSeek1MChat, Code
deepseek/deepseek-v4-flashDeepSeek1MChat
moonshotai/kimi-k2.6MoonshotAI256KChat
moonshotai/kimi-k2.7-codeMoonshotAI256KCode
qwen/qwen3-235b-a22bAlibaba128KChat, Code
meta-llama/llama-4-maverickMeta1MChat, Vision
google/gemma-3-27bGoogle128KChat

View the full, current library — with filtering and live pricing — at tensormachine.ai/models.

API endpoint

All models are accessed via:

POST https://edge.tensormachine.ai/v1/chat/completions

The request format is identical to the OpenAI Chat Completions API.

Choosing a model

ScenarioRecommended model
High-volume chat at low costdeepseek/deepseek-v4-flash
Complex reasoning & codingdeepseek/deepseek-v4-pro
Long document analysis (1M ctx)z-ai/glm-5.2 or minimax/minimax-m3
Image understandingminimax/minimax-m3 or meta-llama/llama-4-maverick
Code generationmoonshotai/kimi-k2.7-code
Multilingual (Hindi, etc.)moonshotai/kimi-k2.6
Budget-friendly open modelgoogle/gemma-3-27b

Model parameters

All standard OpenAI parameters are supported:

ParameterDescription
modelModel ID (required)
messagesConversation history (required)
max_tokensMaximum tokens to generate
temperatureSampling temperature (0–2)
top_pNucleus sampling
streamEnable streaming (SSE)
stopStop sequences
nNumber of completions

Requesting new models

If you need a model not yet on the platform, request it from the console (Models → Request a model). We pool demand and stand up dedicated deployments on a first-come, first-served basis — see how model requests work.

On this page