Source: https://www.tera.gw/inference Retrieved: 2026-09-29T06:17:07.808Z (page retrieval, not inventory verification) Content SHA-256: dfea47ceb3ca875bc7169ffb6ec736599a3e74fdda6c5f9e1fd0169a88ea3884 # Lowest-cost inference. Models you actually use. Open-weight models: Qwen, Llama, DeepSeek, GLM, gpt-oss. OpenAI-compatible. US-processed, zero retention. [Create your API key]() [Get free credits]() [See what's inside ↓]() Lowest cost Per token. No minimums, no contracts. Zero retention Never trained on, never logged. US-based Team, GPUs, and data centers. Drop-in API Hermes Agent, OpenClaw, and any OpenAI-compatible tool. Why Tera 01 ### Lowest-cost open-source inference. Per-token pricing on the models developers actually ship. No minimums, no contracts, no enterprise-tier negotiations. See the rate before you run the request. 02 ### Your data stays yours. Zero retention. Zero training. Zero human review. Prompts and completions are processed and discarded. Backed by a written privacy policy, not a press release. 03 ### Processed entirely in the US. US team. US-owned GPUs. US data centers. Your inference never crosses a border. Lower latency for US customers because the hardware is close to them. 04 ### Open-weight only, by design. We don't have a proprietary model to upsell you to. Tera is a pure execution layer for the open ecosystem. The day a better open model ships, it's on Tera. Integrate ## Drop-in replacement for the OpenAI SDK. Same API surface, lower-cost open-source models. Point your existing client at api.tera.gw and ship. - 01 Get your API key Sign up and grab a key from the dashboard. - 02 Change one line Set base\_url to api.tera.gw and keep the rest of your stack. - 03 Pick a model Qwen, Llama, DeepSeek, GLM, gpt-oss. Python TypeScript Shell api.tera.gw/v1 Copy `````` from openai import OpenAI client = OpenAI( base_url = "https://api.tera.gw/v1" , api_key = "tera_sk_..." ) response = client.chat.completions.create( model = "openai/gpt-oss-120b" , messages = [ { "role" : "user" , "content" : "Hello, Tera." } ] ) print (response.choices[ 0 ].message.content) `````` Models ## Production-ready open models. Unified API. Top-tier open-source models, served from owned U.S. infrastructure with zero data retention. Anthropic claude-fable-5 200K context $10 in · $1 cache read · $50 out Anthropic claude-opus-4-8 200K context $5 in · $0.5 cache read · $25 out Anthropic claude-sonnet-5 200K context $2 in · $0.2 cache read · $10 out DeepSeek DeepSeek-R1-0528 164K context $1.35 in · $5.4 out DeepSeek DeepSeek-V3.1 164K context $0.6 in · $0.06 cache read · $1.7 out DeepSeek DeepSeek-V3.2 164K context $0.56 in · $0.06 cache read · $1.68 out DeepSeek DeepSeek-V4-Flash 1M context $0.19 in · $0.03 cache read · $0.51 out DeepSeek DeepSeek-V4-Pro 1M context $1.74 in · $0.14 cache read · $3.48 out Google gemma-4-26B-A4B-it 262K context $0.15 in · $0.01 cache read · $0.6 out Meta Llama-3.3-70B-Instruct 128K context $0.72 in · $0.72 out MiniMax MiniMax-M2 197K context $0.3 in · $0.03 cache read · $1.2 out Moonshot Kimi-K2.5 262K context $0.6 in · $0.1 cache read · $3 out Moonshot Kimi-K2.6 262K context $0.95 in · $0.16 cache read · $4 out Moonshot kimi-k2-thinking 262K context $0.6 in · $0.06 cache read · $2.5 out Moonshot AI Kimi-K3 1M context $3 in · $0.3 cache read · $15 out OpenAI gpt-5.5 1M context $5 in · $0.5 cache read · $30 out OpenAI gpt-5.6-luna 1M context $0.2 in · $0.02 cache read · $1.2 out OpenAI gpt-5.6-sol 1M context $5 in · $0.5 cache read · $30 out OpenAI gpt-5.6-terra 1M context $2 in · $0.2 cache read · $12 out OpenAI gpt-oss-120b 131K context $0.09 in · $0.36 out OpenAI gpt-oss-20b 131K context $0.07 in · $0.25 out Qwen / Alibaba Qwen3-235B-A22B-Instruct-2507 262K context $0.22 in · $0.88 out Qwen / Alibaba Qwen3-Coder-480B-A35B-Instruct 262K context $0.22 in · $0.02 cache read · $1.8 out Qwen / Alibaba Qwen3-Next-80B-A3B-Instruct 262K context $0.15 in · $1.2 out Qwen / Alibaba Qwen3-Next-80B-A3B-Thinking 262K context $0.15 in · $1.2 out Z.ai GLM-4.7 200K context $0.6 in · $2.2 out Z.ai GLM-5 200K context $1 in · $0.1 cache read · $3.2 out Z.ai GLM-5.2 262K context $1.49 in · $0.27 cache read · $4.62 out Full model list and pricing at [/pricing](). Built for agent workloads ## Everything agent builders actually need. Designed for the inference patterns that agents generate: streaming, tool calls, reasoning traces, and high call volume. OpenAI-compatible tool calling The tools API works without modification. Any framework that uses OpenAI function calling routes through Tera unchanged. Streaming responses Server-sent events on every model. Low time-to-first-token so your agents stay responsive under load. Reasoning content Chain-of-thought traces surfaced as a dedicated field on reasoning models. Same API shape, no custom parsing. Not billed for failed requests 5xx errors and rate-limit responses (429) are not charged. You pay only for tokens that are successfully processed. No platform fee Per-token pricing with no minimums, no contracts, and no monthly base charge. See the rate before you run the request. Works with Hermes Agent, OpenClaw, Cline, and any OpenAI-compatible framework. FAQ ## Common questions. ### Is Tera's API OpenAI-compatible? Yes. Change one line: set base\_url (or baseURL) to https://api.tera.gw/v1. The rest of your OpenAI SDK code works unchanged. Chat completions, streaming, and tool calling are all supported. ### What models are available? Qwen, Llama, DeepSeek, GLM, gpt-oss, MiniMax, and Gemma, with more added as production-quality open models ship. The full list with per-token pricing is at tera.gw/pricing. ### How do I get started? Sign up at tera.gw/dashboard, create an API key, and you can make your first call within minutes. No approval process, no contracts, no minimums. ### What does Tera cost? Per-token pricing, no platform fee, no minimums. Prices vary by model. See the full rate card at tera.gw/pricing. Failed requests (5xx errors, 429 rate limits) are not billed. ### Where is my data processed? All inference runs on US-owned infrastructure in US data centers. Prompts and completions are processed and discarded. Zero logging, zero retention, zero training on your data. ### Does Tera support streaming and tool calling? Yes. Set stream: true for server-sent events on any model. The OpenAI tools API (function calling) is supported. Reasoning models also surface chain-of-thought traces in a dedicated reasoning field. ### What agent frameworks work with Tera? Any framework that speaks the OpenAI API works without modification: LangChain, LlamaIndex, AutoGen, CrewAI, and tools like Hermes Agent, OpenClaw, and Cline. Self-serve ## Direct API access. Live now. Sign up, get a key, and start running open-weight models today. No waitlist, no approval process. Pay per token, no minimums. [Create your API key]() [Already have an account? Sign in →]() --- Interactive controls above are described, not submitted. Reading does not reserve capacity or authorize payment. For GPU requirements: https://www.tera.gw/gpu-rentals#capacity-brief