Unified API for
LLM inference.
Every leading LLM behind one drop-in OpenAI-compatible API. Open source LLMs run on Runware's own infrastructure at up to 80% lower cost.
Refactor this Python function for readability.
Ship faster. Spend less.
Run LLM inference on infrastructure built for production. Leave reliability, scale, and operations to us.
Open source and closed source, same endpoint.
Leading closed source LLMs from Anthropic, OpenAI, Google and xAI alongside the best open source LLMs running on Runware's own hardware. Pick by quality, cost or context window. Same auth either way.
Top open source LLMs on custom hardware.
Leading closed source LLMs at market leading rates.
Match the model to the job.
Every workload has a model that suits it. Pick a task to see what we'd reach for.
Coding agents & engineering
Refactor at codebase scale, generate tests, run autonomous tickets end to end. Models here lead public SWE-bench Verified and Pro deployments, with strong tool calling and long-horizon execution.
- Try in PlaygroundGLM-4.7Open~9x cheaper than Sonnet 4.6zai:[email protected]
Best open source coder. 73.8% on SWE-bench Verified at a fraction of closed source pricing.
- Try in PlaygroundClaude Sonnet 4.6Closedanthropic:[email protected]
Production default for coding agents. Strong on branched, multi-step tickets and computer-use scaffolds.
- Try in PlaygroundMiniMax M2.7Open~13x cheaper than Sonnet 4.6minimax:m2.7@0
Reliable agentic coding with strong instruction-following on long-horizon tool workflows.
Open source has closed the gap.
On most production workloads, open source LLMs now match closed source on quality, at a fraction of the cost. On some, they're the strongest available choice.
358B MoE with interleaved thinking. Scores 73.8% on SWE-bench Verified at a fraction of closed source pricing.
Sparse MoE at 397B total / 17B active. 262K native context, extensible to ~1M. Built for long-context reasoning at scale.
Top open source option for multimodal agents. Image and video understanding alongside long-horizon software tasks.
Long-context agentic coding tuned for production tool use. Holds quality at high throughput.
Intelligence: SWE-bench Verified scores from public May 2026 model reports. Cost: Runware list price per million output tokens, pulled live from the model catalog. Open source models run on Runware's own hardware; closed source passed through at market leading rates.
What you'd save by switching to open source.
Pick a workload and a monthly token volume. We'll substitute the closed source model you're using today with an equivalent open source model on Runware, and show the monthly saving at your scale.
Comparison against the equivalent closed source model billed direct from the provider. Real Runware list prices, blended across a representative input/output mix for the selected workload.
Already on the OpenAI SDK? Change two values.
The Runware text endpoint speaks the OpenAI protocol. The request shape, streaming format and SDKs you already use carry over unchanged. Point your existing client at https://api.runware.ai/v1, swap the API key and pick a Runware model.
from openai import OpenAI
client = OpenAI(
api_key="your_runware_api_key",
base_url="https://api.runware.ai/v1",
)
response = client.chat.completions.create(
model="minimax:m2.7@0",
messages=[{"role": "user", "content": "What is the capital of France?"}],
max_completion_tokens=256,
)
print(response.choices[0].message.content)"stream": true. Identical SSE format. Use the chunks you already parse.
delta.reasoning_content streams before delta.content on models that support it.
Use the native /v1 endpoint for taskUUID, async delivery, includeCost, and one shape across every modality.
LLM inference built for demanding products.
Your data stays yours.
FAQ
Get an API key.
Pick any model on Runware. The OpenAI SDK works as-is, or use our native API for more control.