GLM 5.21M contextZ.ai
Large-scale reasoning model with a 1M-token context for long-horizon agents and project-level engineering.
Chompute Inference · Under the AI Control Plane
Top open-weight models. The cheapest way to run them.
Kimi, GLM, DeepSeek, GPT OSS, MiniMax, and Qwen behind one gateway that speaks your existing OpenAI or Anthropic SDK. Prepaid, so spend can never run past budget. Pay per token — nothing else.
Kimi K2.7 Code262K contextMoonshot AI
Coding-focused agentic model built for reliable end-to-end software engineering across long contexts.
DeepSeek V4 Pro1M contextDeepSeek
Flagship MoE reasoning model for advanced coding, long-context analysis, and complex agent workflows.
MiniMax M3512K contextMiniMax
Sparse-attention multimodal model suited to coding agents, document-heavy workflows, and multi-step reasoning.
Kimi K2.6262K contextMoonshot AI
Native multimodal agent model for long-horizon coding, UI generation, tool use, and agentic workflows.
Qwen3.7 Plus262K contextAlibaba
Cost-effective multimodal agent model for coding, screenshot-driven workflows, tool use, and planning.
DeepSeek V4 Flash1M contextDeepSeek
Efficiency-optimized MoE model for fast, cost-effective reasoning, coding, and agents over 1M-token contexts.
GPT OSS 120B131K contextOpenAI
OpenAI's open-weight MoE model for high-reasoning, agentic, and production workloads with function calling.
$5
Is all it takes to start
prepaid balance — no subscription, no commitments
$0
Platform fee
you pay for tokens, nothing else
3
API surfaces
OpenAI Chat, OpenAI Responses, Anthropic Messages
1M
Tokens of context
on the longest-context models in the catalog
Live catalog
The models teams are reaching for.
A curated lineup of the best open weights — including these, live right now. New frontier releases land here days after they ship, with per-token pricing shown in the console.
Kimi K2.6
Moonshot AINative multimodal agent model for long-horizon coding, UI generation, tool use, and agentic workflows.
262K contextVision
Kimi K2.7 Code
Moonshot AICoding-focused agentic model built for reliable end-to-end software engineering across long contexts.
262K contextVision
GLM 5.2
Z.aiLarge-scale reasoning model with a 1M-token context for long-horizon agents and project-level engineering.
1M context
DeepSeek V4 Flash
DeepSeekEfficiency-optimized MoE model for fast, cost-effective reasoning, coding, and agents over 1M-token contexts.
1M context
DeepSeek V4 Pro
DeepSeekFlagship MoE reasoning model for advanced coding, long-context analysis, and complex agent workflows.
1M context
GPT OSS 120B
OpenAIOpenAI's open-weight MoE model for high-reasoning, agentic, and production workloads with function calling.
131K context
MiniMax M3
MiniMaxSparse-attention multimodal model suited to coding agents, document-heavy workflows, and multi-step reasoning.
512K contextVisionVideo
Qwen3.7 Plus
AlibabaCost-effective multimodal agent model for coding, screenshot-driven workflows, tool use, and planning.
262K contextVision
And the list keeps growing. New frontier open weights go live days after release — you never wait on a roadmap to run the latest model.
Drop-in
Three lines to switch. Keep the rest of your code.
The gateway is compatible with the APIs you already use. Point your SDK at your Chompute base URL, add your access key, and pick a model — streaming, tool calling, and structured outputs included.
- ✓OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages
- ✓Drop-in over your existing SDK — no rewrite, no new dependencies
- ✓Streaming, function/tool calling, and structured outputs supported
- ✓Reasoning controls per model — thinking budgets, effort levels
◇ OpenAI SDK · Python
01from openai import OpenAI0203client = OpenAI(04 base_url="YOUR_GATEWAY_URL/v1", # shown in your console05 api_key="unused", # auth is the header below06 default_headers={"X-Chompute-Access-Key": "YOUR_ACCESS_KEY"},07)0809stream = client.chat.completions.create(10 model="kimi-k2.7-code",11 messages=[{"role": "user", "content": "Refactor this handler."}],12 stream=True,13)1415for chunk in stream:16 print(chunk.choices[0].delta.content or "", end="")The economics
Why this is the cheapest path.
Open-weight models rewrote the price of intelligence. Chompute is built to pass that price through, not mark it up.
Open weights closed the gap.
The best open-weight models now sit next to closed frontier models on the benchmarks that matter — coding, tool use, long-horizon agents — at a fraction of the per-token price.
The cheapest way to run them.
Chompute serves top open-weight models at the lowest per-token cost. Live pricing for every model is in the console — the price you see is the price you pay.
Pay for tokens. Nothing else.
No platform fee, no seats, no minimum commitment. Your balance only moves when you call the API.
Budget certainty
Spend that can't surprise you.
Prepaid is not a limitation — it is the point. Your balance is a hard ceiling that no agent, no team, and no runaway loop can break through.
Prepaid by design.
Top up $5, $50, or $100. Requests draw down the balance, and spending stops at zero — you can never go over budget. No card on file quietly running up an invoice.
Every token attributed.
Live usage insights in the console — requests, tokens, and cost per model — so you always know exactly where spend is going.
Keys per team or project.
Issue separate API keys per team, app, or environment — and track usage against each key.
How it starts
Sign in to streaming tokens in minutes.
01
Sign in
Google sign-in and you are in the console. No sales call, no procurement cycle, no card required to look around.
02
Add balance
Start with $5 through Stripe checkout. Your balance is your hard budget ceiling — top up only when you choose to.
03
Point your SDK
Swap the base URL, add your access key, keep your code. Your first tokens stream back in minutes.
When you scale
Starts like a tool. Governs like a platform.
Chompute Inference runs under the Chompute AI Control Plane. The endpoint one developer adopts in an afternoon carries the attribution, budgets, and audit trail an organization needs later — nothing to migrate.
Under the control plane
- ✓Same access layer and audit trail as every governed AI surface
- ✓Usage attribution ready for finance and platform reviews
- ✓Concurrency and spend controls per account
- ✓Scales from one developer to org-wide policy on the control plane
FAQ
Questions, answered.
Will it work with my existing code?
Yes. The gateway speaks OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages. You point your existing SDK at your Chompute base URL, add your access key as a header, and keep the rest of your code — streaming and function/tool calling included.
Which models do I get?
The best open-weight models from the labs pushing the frontier — Moonshot, Z.ai, DeepSeek, OpenAI, MiniMax, and Alibaba among them — with new releases added as they land. You enable or disable models per account, and the live catalog in the console is always the current source of truth.
How does the prepaid billing work?
You top up your balance — $5, $50, or $100 — through secure Stripe checkout. Every request draws down the balance at the per-token price shown live in the console. There is no subscription and no platform fee; your balance is the only money involved.
What happens if my balance runs out?
Requests stop with a clear error and nothing else is billed — there is no overage, ever. Top up and the same key starts working again immediately.
Why not just show prices on this page?
Per-token prices move as we push costs down, so the console shows live pricing for every model before you spend anything. What does not move: the promise that this is the cheapest way to run top open-weight models.
Can I create multiple API keys?
Today each account gets one access key. Separate keys per team, app, or environment — each with its own usage tracking — are on the roadmap and coming soon.
What about enterprise volume or governance?
Chompute Inference runs under the Chompute AI Control Plane, so the same attribution, budgets, and policy machinery that governs enterprise AI estates applies here. For volume terms or a governed rollout, write to sales@chompute.ai.
Get started
Run top open-weight models in minutes.
Sign in, add $5, point your SDK. No subscription, no platform fee, no way to go over budget.