Wafer
on Orq.ai
Use Wafer’s ultra‑fast open‑source LLMs through a single Orq.ai integration. Route models such as GLM‑5.2, GLM‑5.1, Kimi‑K2.6/K2.7 Code, Qwen3.5‑397B, Qwen3.6‑35B, Qwen3.7‑Max, DeepSeek V4 Pro/Flash, MiniMax M3, and other Turbo models via Orq’s AI Router for chat, reasoning, coding, and agentic workloads.
Capabilities:
Chat
Reasoning
Vision
Models Supported:
GLM-5.2
GLM5.2-Fast
GLM-5.1
Kimi-K2.6
Qwen3.5-397B-A17B
Provider HQ:
Wafer, a US‑based AI platform focused on high‑throughput open‑source inference
Access Together AI through Orq.ai’s AI Router
Wafer is an AI inference optimisation platform and model host that focuses on making open‑source LLMs fast and predictable for production, with OpenAI‑compatible and Anthropic‑compatible endpoints. Wafer Serverless gives pay‑as‑you‑go access to Wafer‑hosted models via one inference endpoint, while Wafer Pass subscriptions offer flat‑rate, near‑unlimited usage for Turbo models.
Wafer models available on Orq.ai
Model | Type | Context / scope | Best for | Pricing tier |
|---|---|---|---|---|
GLM‑5.2 | Chat / reasoning / coding | Around 200K+ context | Flagship GLM‑5.2 with strong coding and reasoning; high‑stakes tasks needing more depth | Premium – $1.20 input / $4.10 output, cache around $0.20 / 1M tokens |
GLM‑5.1, Qwen3.5‑397B | Chat / reasoning / tools | ~200K–262K context | Everyday production workloads, RAG, and general assistants with strong quality vs cost | Mid‑tier – GLM‑5.1 at $1.00 in / $3.20 out, Qwen3.5‑397B at $0.43 in / $2.60 out |
DeepSeek V4 Flash, Qwen3.6‑35B, MiniMax M3 | Chat / fast / cost‑efficient | Up to ~256K–1M context | Fast, cheaper tasks such as high‑volume chat, classification, extraction, and agents | Cost‑efficient – DeepSeek V4 Flash at $0.09–0.14 in / $0.18–0.28 out, Qwen3.6‑35B at $0.15 in / $1.00 out, MiniMax M3 at $0.33 in / $1.32 out |
Why use Wafer through Orq.ai?
Capability | Provider / models | Direct | Through Orq.ai |
|---|---|---|---|
Chat & reasoning | GLM‑5.2, GLM‑5.1, Qwen3.5, Qwen3.7‑Max, MiniMax M3 | Call Wafer’s OpenAI‑compatible pass.wafer.ai/v1 endpoint directly for chat and reasoning.docs | Use Wafer models through Orq.ai’s router, adding central routing, tracing, evals, budgets, and governance around each request |
Coding | DeepSeek V4 Pro/Flash, GLM‑5.x, Qwen‑coder variants | Use Wafer directly for high‑throughput coding agents, debugging, and code‑aware assistants | Route coding workloads through Orq.ai, compare Wafer‑served models against other providers, and monitor cost, latency, and quality from one control layer |
High‑throughput agents | DeepSeek V4 Flash, GLM‑5.x Turbo, Qwen Turbo | Use Wafer Pass subscriptions with near‑unlimited requests for continuous agents | Use Orq.ai to send only your agentic, throughput‑heavy workloads to Wafer while routing other tasks to different providers, all under unified observability |
This gives teams a practical way to use Wafer where it performs best while centralising routing, observability, evals, and cost controls across the wider model stack.
Pricing
Plans and API access
Wafer, a US‑based AI platform focused on high‑Wafer offers two main ways to pay:everydev+2
Serverless (pay‑as‑you‑go): pay per token via OpenAI‑compatible https://pass.wafer.ai/v1 and Anthropic‑compatible endpoints, funding usage with credits and no subscription window.
Wafer Pass subscriptions: flat weekly or monthly pricing with request caps per 5‑hour window, covering all Turbo models with zero data retention on certain plans.
Representative serverless token pricing (per 1M tokens):
GLM‑5.2: $1.20 input / $4.10 output / $0.20 cache.
GLM‑5.1: $1.00 input / $3.20 output / $0.10 cache.
Kimi‑K2.6: $0.68 input / $3.15 output / $0.07 cache.
DeepSeek V4 Pro: $1.20–1.74 input / $2.40–3.48 output / low cache.
DeepSeek V4 Flash: $0.09–0.14 input / $0.18–0.28 output / $0.01 cache.
Qwen3.5‑397B: $0.43 input / $2.60 output / $0.04 cache.
Qwen3.6‑35B: around $0.15 input / $1.00 output.
Qwen3.7‑Max: $5.00 input / $15.00 output.
Representative Wafer Pass subscription pricing (Turbo models, approximate weekly equivalents):
Starter: about $10/week, 1,000 requests per 5‑hour window, all Turbo models, OpenAI+Anthropic‑compatible API.
Privacy / Pro: about $25/week, 2,000–5,000 requests per 5‑hour window, all Turbo models, Zero Data Retention on privacy tiers.
Max: about $50–63/week, 20,000 requests per 5‑hour window, all Turbo models plus priority routing.
Every subscription tier includes every current Turbo model and future Turbo models, with no price increase as the catalog grows.
Check Wafer’s pricing page and your Orq.ai provider configuration for current rates and to decide between pay‑as‑you‑go vs Wafer Pass for your workloads.docs.orq+1throughput open‑source inference.
Compatible frameworks and tools
Wafer exposes:
OpenAI‑compatible endpoint: https://pass.wafer.ai/v1
Anthropic‑compatible endpoint: https://pass.wafer.ai/v1/messages
Auth via Authorization: Bearer <key>, with optional per‑request Zero Data Retention header Wafer-ZDR: required.
Orq.ai exposes Wafer through:
a Wafer provider configuration in AI Gateway / Integrations, where you paste your Wafer API key and optionally set ZDR policies, and
an OpenAI‑compatible routing layer, so existing OpenAI SDK workflows can target Wafer via Orq’s base URL.
That means:
Existing backend workflows using OpenAI SDKs or Anthropic‑style clients can switch to Orq’s router and route some traffic to Wafer without changing request payloads.
Agents, code assistants, and tools that integrate with Orq.ai can be configured so that high‑throughput or privacy‑sensitive steps are served by Wafer, while other steps use different providers, all sharing the same observability and governance layer.
Check the Orq.ai integration docs for the latest supported frameworks and tools for Wafer.
FAQs
Do I need a separate Wafer account to use Wafer through Orq.ai?
Yes. You create a Wafer account, load credits or choose a Wafer Pass plan, mint an API key, then configure it in Orq.ai’s AI Gateway; Orq.ai then gives you one place to manage routing, observability, and cost controls around that Wafer usage.
Can I route only some workflows to Wafer and others to different providers?
Yes. You define routes per workflow in Orq.ai and choose which ones should use Wafer vs other providers, so you can reserve Wafer for high‑throughput open‑source workloads while sending other tasks to search‑grounded, proprietary, or regional models.
Does using Wafer through Orq.ai add latency?
Orq.ai is designed as a lightweight router, so the added overhead is small compared to model inference time; you can use routing policies, caching, and provider selection to keep end‑to‑end performance within your targets while gaining visibility and control.
Alternatives to
Wafer
Anthropic
Use Claude Opus 4.7, Sonnet 4.6, Haiku 4.5, and other supported Claude models through one API.
Chat
Code
Reasoning
Vision
Models:
claude-opus-5
claude-sonnet-5
claude-fable-5
Open AI
Use OpenAI's foundation models through a single Orq.ai API. Route models such as GPT-4.1, GPT-4.1-mini, o3-mini, and GPT-4o-class models via Orq's AI Router for chat, reasoning, coding, and multimodal workloads.
Chat
Code
Image Generation
Reasoning
Speech
Vision
Models:
gpt-5.5 (EU)
gpt-5.6-luna (EU)
gpt-5.6-sol (EU)
Google AI
Use Google’s Gemini models through a single Orq.ai API. Route models such as Gemini 3.1 Pro, Gemini 2.5 Flash, and Gemini 2.0 Flash‑Lite via Orq’s AI Router for chat, reasoning, coding, and multimodal workloads.
Chat
Code
Embeddings
Image Generation
Reasoning
Speech
Vision
Models:
gemini-3.5-flash-lite (Gemini API)
gemini-3.6-flash (Gemini API)
Gemini 3 Pro Image
AWS
Use AWS Bedrock’s foundation models through a single Orq.ai API. Route models such as Amazon Nova, Amazon Titan Text, and compatible third‑party models exposed via Bedrock through Orq’s AI Router for chat, reasoning, coding, and multimodal workloads.
Chat
Code
Embeddings
Reasoning
Vision
Models:
eu.anthropic.claude-opus-5
global.anthropic.claude-opus-5
us.anthropic.claude-opus-5
