Thinking Machines vs StreamLake

StreamLake, Kuaishou's AI cloud, sells KAT-Coder, a proprietary agentic coding model. Thinking Machines offers open Inkling weights and a training API to build your own specialist.

By The Subconscious Team · Updated September 30, 2026

Thinking Machines vs StreamLake: key differences

StreamLake's pitch is a ready coding model. KAT-Coder-Pro V2.5, from Kuaishou's KwaiKAT team, was trained with large-scale agentic RL for repository-level work, according to StreamLake: reading issues, editing across files, running tests and fixing its own errors. It is sold per token or through a KwaiKAT Coding Plan, with OpenAI-protocol endpoints and a Claude-protocol proxy for Claude Code. Thinking Machines offers the ingredients rather than a finished coder. Tinker lets teams run their own RL loops on open bases like Qwen3.5, GLM-5.3 or Kimi K2.6, and Inkling ships under Apache 2.0 with 1M context.

Openness and residency set them apart. KAT-Coder is proprietary, and StreamLake leads with China, yuan pricing and China-hosted data, which rules it out for many US and EU buyers. Thinking Machines is US based and its Inkling weights can be self-hosted anywhere. StreamLake also sells bare-metal compute, while Thinking Machines' serving is a beta API for Inkling at $1.00 in and $4.05 out plus a checkpoint endpoint limited to testing. For a developer who wants a cheap coding agent today, StreamLake's plan is simpler. For a team training its own agentic coder, Tinker gives more control.

What Thinking Machines and StreamLake do

Thinking Machines

Thinking Machines Lab is the San Francisco lab Mira Murati, formerly CTO of OpenAI, founded in February 2025, with OpenAI co-founder John Schulman as chief scientist. It raised about $2 billion at a $12 billion valuation in July 2025 in a round led by Andreessen Horowitz, and in March 2026 signed a multi-year Nvidia deal for one gigawatt of Vera Rubin capacity. Its main developer product is Tinker, an API for post-training open-weight models that launched in October 2025 and is now generally available. Tinker exposes four low-level calls, forward_backward, optim_step, sample and save_state, so teams write their own supervised or reinforcement learning loops while Thinking Machines runs the distributed GPU work. Training uses LoRA adapters rather than full weight updates.

Example models: Inkling, Inkling-Small, Qwen3.5, Kimi K2.6

Full Thinking Machines profile

StreamLake

StreamLake is the AI cloud brand of Kuaishou, the Chinese short-video company behind the Kling video models. It sells model-as-a-service inference and bare-metal compute to internet businesses, drawing on the infrastructure Kuaishou built to serve video at massive scale. Its developer site offers APIs, SDKs and integration guides aimed at taking teams from testing to production.

Example models: KAT-Coder-Pro V2.5, KAT-Coder-Air

Full StreamLake profile

Should you choose Thinking Machines or StreamLake?

Thinking Machines

Choose Thinking Machines for

Training your own agentic coding model with RL
Open weights for self-hosting outside China
Multimodal input with 1M context

StreamLake

Choose StreamLake for

Subscription access to a coding model
A Claude Code-compatible coding endpoint
Chinese businesses needing domestic MaaS

Thinking Machines vs StreamLake at a glance

Attribute	Thinking Machines	StreamLake
Model access	Open weights	Proprietary coding models
Flagship models	Inkling, Inkling-Small	KAT-Coder-Pro V2.5, KAT-Coder-Air
Speed	Unknown	Unknown
Price	Per 1M tokens by prefill, sample, train; Inkling $1.00 in, $4.05 out	Per token or KwaiKAT Coding Plan
Customization	LoRA SFT and RL via Tinker	Unknown
Deployment	Training API, beta serverless (Inkling only)	MaaS API, bare metal
Long context	Inkling up to 1M; Tinker 32K–256K	Unknown

Frequently asked questions

What is the difference between Thinking Machines and StreamLake?

StreamLake, Kuaishou's AI cloud, sells KAT-Coder, a proprietary agentic coding model. Thinking Machines offers open Inkling weights and a training API to build your own specialist.

When should I choose Thinking Machines over StreamLake?

Training your own agentic coding model with RL; Open weights for self-hosting outside China; Multimodal input with 1M context.

When should I choose StreamLake over Thinking Machines?

Subscription access to a coding model; A Claude Code-compatible coding endpoint; Chinese businesses needing domestic MaaS.

Is Thinking Machines or StreamLake cheaper?

Thinking Machines: Per 1M tokens by prefill, sample, train; Inkling $1.00 in, $4.05 out. StreamLake: Per token or KwaiKAT Coding Plan. The cheaper choice depends on the model and workload.

Related comparisons

Running long-horizon agents?

If your agents run past 200K tokens, compare both against Subconscious. Our inference stack treats a long-horizon trace as the primary workload, so speed, cost, and accuracy hold up deep into the trace.

Subconscious vs Thinking Machines Subconscious vs StreamLake