Inference for Coding Agents
Frontier coding agents, 100% in your cloud
We run the powerful GLM-5.2 model on our inference system, so you can run coding agents securely in your cloud. Your engineers use a top-tier model, with all the benefits of an on-prem deployment.
Your cloud, powering the harnesses your engineers love.
OpenCode
Claude Code
Codex
Cursor
Pi
ZedUnlimited frontier coding agents
For the first time, own the frontier of AI
For the first time an open model, GLM-5.2, matches Opus and GPT on coding ability. Build with frontier ability all in your cloud, and get a stack of benefits no closed API can match.
PRIVACY
Your data stays yours
Your data and privacy are 100% guaranteed. Code and prompts stay under your control and never train anyone else’s model.
UNFILTERED
Powerful and unrestricted
A fully controlled implementation gives you an unrestricted model, with no refusals or silent performance degradation getting in the way of legitimate work.
RELIABILITY
You control uptime
Your engineers rely on coding agents, but Opus and GPT hit constant rate limits and server issues. Run in your own environment and your engineers never wait on someone else’s outage.
Powered by Subconscious
Your cloud, with our inference system
Our inference system is the most efficient way to run GLM-5.2 on your own GPUs for coding agents. We designed it for agents from the start, so the same hardware does far more work.
CONCURRENCY
Half as many GPUs
Our system runs 2.3x the concurrent workloads of generic serving, so your team needs half as many GPUs to handle the same work.
THROUGHPUT
Faster token throughput
3.5x faster token throughput, sustained even deep down the long reasoning chains where generic serving slows.
CONTEXT
Longer context reasoning
10x stateful context extension, so agents hold the thread across big repos and long sessions instead of degrading mid-run.
Don't take our word for it
Visualize Subconscious GLM-5.2 vs. Claude Opus 4.8
We tested the models on seven identical coding prompts. See the artifacts for yourself.
Claude Opus 4.8
Subconscious GLM-5.2
Simple migration
Change the base URL. Keep on coding.
No new SDK, no rewritten prompts, no different tool format. Your agent keeps the harness it has today and picks up a new model behind it.
# Claude Code
$export ANTHROPIC_BASE_URL=https://api.subconscious.dev
$export ANTHROPIC_AUTH_TOKEN=<your API key>
# OpenCode, or anything on the OpenAI SDK
$export OPENAI_BASE_URL=https://api.subconscious.dev/v1
$export OPENAI_API_KEY=<your API key>
OpenCodeClaude CodeCodexCursorPi
ZedTrial with our API
Open models, served via API
Point your coding agent at our API and run open-source models on our inference system, at a fraction of frontier pricing.
TIM-Qwen3.6 27B
subconscious/tim-qwen3.6-27b
Post-trained on Qwen3.6 27B, served on our TIMRUN runtime. Natively multimodal, accepting both text and images. An extremely efficient and capable model for agentic workflows.
$0.30 input, $0.15 cached input, $3.00 output, per 1M tokens.
GLM-5.2
subconscious/glm-5.2
GLM-5.2, served on our TIMRUN runtime and built for agentic coding and long-running agent work. Text-only input.
$1.40 input, $0.26 cached input, $4.40 output, per 1M tokens.
Dedicated deployment
A private inference system for your engineers
Run frontier coding agents, in your own cloud, with your own hardware. Privacy, security, and unlimited usage.
Chat about a dedicated deployment for my company →“It was faster than a lot of LLMs I've used, and it was easy to set up.”
Cedric Prentice
Software Engineer @ Wayfair
“Made the whole process very easy!”
Smruthi Ramesh
Lead Data Scientist @ Schneider Electric
“It was great!”
Kevin Sullivan
Director @ EY-Parthenon
“Very cool, easy to use!”
Bill Simmons
Co-Founder @ Orbit.me
“It was really fast!!”
Inder Singh
UDE
“Easy to use, great UI”
Yassine Fatimi
Founder @ ClauseGuard
“Pretty easy to use. No brainer. Easy drop in for OpenAI.”
Hansen Liang
Founder @ stealth
“It worked very well and provided accurate descriptions of the images we passed in. The API costs were very cheap.”
Sam Mayle
Researcher @ Mitsubishi Electric Research Lab
“It was faster than a lot of LLMs I've used, and it was easy to set up.”
Cedric Prentice
Software Engineer @ Wayfair
“Made the whole process very easy!”
Smruthi Ramesh
Lead Data Scientist @ Schneider Electric
“It was great!”
Kevin Sullivan
Director @ EY-Parthenon
“Very cool, easy to use!”
Bill Simmons
Co-Founder @ Orbit.me
“It was really fast!!”
Inder Singh
UDE
“Easy to use, great UI”
Yassine Fatimi
Founder @ ClauseGuard
“Pretty easy to use. No brainer. Easy drop in for OpenAI.”
Hansen Liang
Founder @ stealth
“It worked very well and provided accurate descriptions of the images we passed in. The API costs were very cheap.”
Sam Mayle
Researcher @ Mitsubishi Electric Research Lab
“It was faster than a lot of LLMs I've used, and it was easy to set up.”
Cedric Prentice
Software Engineer @ Wayfair
“Made the whole process very easy!”
Smruthi Ramesh
Lead Data Scientist @ Schneider Electric
“It was great!”
Kevin Sullivan
Director @ EY-Parthenon
“Very cool, easy to use!”
Bill Simmons
Co-Founder @ Orbit.me
“It was really fast!!”
Inder Singh
UDE
“Easy to use, great UI”
Yassine Fatimi
Founder @ ClauseGuard
“Pretty easy to use. No brainer. Easy drop in for OpenAI.”
Hansen Liang
Founder @ stealth
“It worked very well and provided accurate descriptions of the images we passed in. The API costs were very cheap.”
Sam Mayle
Researcher @ Mitsubishi Electric Research Lab
“It was faster than a lot of LLMs I've used, and it was easy to set up.”
Cedric Prentice
Software Engineer @ Wayfair
“Made the whole process very easy!”
Smruthi Ramesh
Lead Data Scientist @ Schneider Electric
“It was great!”
Kevin Sullivan
Director @ EY-Parthenon
“Very cool, easy to use!”
Bill Simmons
Co-Founder @ Orbit.me
“It was really fast!!”
Inder Singh
UDE
“Easy to use, great UI”
Yassine Fatimi
Founder @ ClauseGuard
“Pretty easy to use. No brainer. Easy drop in for OpenAI.”
Hansen Liang
Founder @ stealth
“It worked very well and provided accurate descriptions of the images we passed in. The API costs were very cheap.”
Sam Mayle
Researcher @ Mitsubishi Electric Research Lab
“It's awesome”
Sam Xifaras
Software Engineer @ Stripe
“Fantastic”
Atin Tandon
Senior Software Engineer @ Sunrun
“Holy f**k it's fast”
Mike Miner
Founder @ Rivilo
“Great stuff”
Wes Donohoe
CEO @ Visitrecall
“Very interesting alternative to OpenAI”
Dave Gogi
Founder @ Signal X
“Pretty fast and fun to use”
Nihir Kothari
Founder @ Sidekick Software
“Very good”
Yikun Ding
CPO @ Firelights Quant
“It's awesome”
Sam Xifaras
Software Engineer @ Stripe
“Fantastic”
Atin Tandon
Senior Software Engineer @ Sunrun
“Holy f**k it's fast”
Mike Miner
Founder @ Rivilo
“Great stuff”
Wes Donohoe
CEO @ Visitrecall
“Very interesting alternative to OpenAI”
Dave Gogi
Founder @ Signal X
“Pretty fast and fun to use”
Nihir Kothari
Founder @ Sidekick Software
“Very good”
Yikun Ding
CPO @ Firelights Quant
“It's awesome”
Sam Xifaras
Software Engineer @ Stripe
“Fantastic”
Atin Tandon
Senior Software Engineer @ Sunrun
“Holy f**k it's fast”
Mike Miner
Founder @ Rivilo
“Great stuff”
Wes Donohoe
CEO @ Visitrecall
“Very interesting alternative to OpenAI”
Dave Gogi
Founder @ Signal X
“Pretty fast and fun to use”
Nihir Kothari
Founder @ Sidekick Software
“Very good”
Yikun Ding
CPO @ Firelights Quant
“It's awesome”
Sam Xifaras
Software Engineer @ Stripe
“Fantastic”
Atin Tandon
Senior Software Engineer @ Sunrun
“Holy f**k it's fast”
Mike Miner
Founder @ Rivilo
“Great stuff”
Wes Donohoe
CEO @ Visitrecall
“Very interesting alternative to OpenAI”
Dave Gogi
Founder @ Signal X
“Pretty fast and fun to use”
Nihir Kothari
Founder @ Sidekick Software
“Very good”
Yikun Ding
CPO @ Firelights Quant