Long-running agents deserve better inference.
Subconscious
vs

Subconscious vs Inception

Inception makes each response fast by generating tokens in parallel with diffusion. Subconscious makes long agent traces fast by changing what the runtime rereads past 200K tokens.

By The Subconscious Team · Updated

Subconscious vs Inception: key differences

Inception's Mercury models are diffusion LLMs: they draft a block of text and refine it in a few passes instead of writing one token at a time. Mercury 2.5 runs at over 1,100 tokens per second by Inception's count, about 640 as measured by Artificial Analysis, costs $0.20 in and $0.75 out, and tops out at 260K context. Its quality sits with cost-optimized models like Claude Haiku 4.5. Subconscious serves open models, GLM 5.3 and DeepSeek V4.1 Flash, on a runtime that prunes the KV cache and preserves suffix state, so long agent traces stop rereading their whole context. It delivers 2x faster task completion, 50% to 80% lower cost than standard inference and a 5M+ effective context window.

The two optimize different things. Inception wins on the single fast response: autocomplete, voice agents, quick extraction. Subconscious wins when one task spans hundreds of steps and millions of tokens, which is past Mercury's 260K window and beyond what a Haiku-class model handles well. Inception's weights are closed; Subconscious runs open models and offers dedicated and on-prem deployments for nearly any open model.

What Subconscious and Inception do

Subconscious

Subconscious is an MIT CSAIL spinout in Kendall Square that builds inference for long-horizon agents, the workloads where a single trace runs past 200K tokens and often into the millions. Its runtime drops in as a replacement for vLLM or SGLang. Instead of rereading an ever-growing context on every step, it prunes the KV cache and preserves suffix state, and Subconscious co-designs the runtime with post-trained model variants it calls Marathon. Against open models on standard inference, Subconscious delivers 2x faster task completion, delivers a 5M+ effective context window, cuts cost 50% and up to 80%, and scores neutral to 10% better on agentic benchmarks.

Example models: GLM 5.3, DeepSeek V4.1 Flash

Full Subconscious profile

Inception

Inception builds diffusion large language models (dLLMs). Every other LLM on this list generates text one token at a time; Inception's Mercury models start from a rough draft of a whole block and refine it over a few passes, producing many tokens in parallel. That ties speed less to output length and lets the models run fast on ordinary NVIDIA GPUs rather than custom chips. The company was founded by professors from Stanford, UCLA and Cornell behind core work on diffusion, Flash Attention and DPO, and is led by Stanford's Stefano Ermon. It raised a $50M seed in November 2025 led by Menlo Ventures, with Mayfield, M12, NVentures, Snowflake and Databricks, plus Andrew Ng and Andrej Karpathy as angels.

Example models: Mercury 2.5, Mercury 2

Full Inception profile

Should you choose Subconscious or Inception?

Subconscious

Choose Subconscious for

  • Coding and research agents past 200K tokens
  • 5M+ effective context on open models
  • Billing on processed tokens rather than tokens sent

Inception

Choose Inception for

  • Code autocomplete and edits that must feel instant
  • Voice agents with time to first token under 170ms
  • Cheap, fast summarization and extraction

Subconscious vs Inception at a glance

AttributeSubconsciousInception
Model accessOpen weightsClosed weights (Mercury dLLMs)
Flagship modelsGLM 5.3, DeepSeek V4.1 FlashMercury 2.5, Mercury 2
Speed2x faster task completion1,100+ tok/s (vendor); ~640 measured
Price50–80% lower cost; billed on processed tokens$0.20 in, $0.75 out (Mercury 2.5)
CustomizationMarathon post-trained variantsFine-tuning on request
DeploymentManaged API, dedicated, on-premServerless API, Bedrock, Azure, private
Long context5M+ effective context260K

Frequently asked questions

What is the difference between Subconscious and Inception?

Inception makes each response fast by generating tokens in parallel with diffusion. Subconscious makes long agent traces fast by changing what the runtime rereads past 200K tokens.

When should I choose Subconscious over Inception?

Coding and research agents past 200K tokens; 5M+ effective context on open models; Billing on processed tokens rather than tokens sent.

When should I choose Inception over Subconscious?

Code autocomplete and edits that must feel instant; Voice agents with time to first token under 170ms; Cheap, fast summarization and extraction.

Is Subconscious or Inception cheaper?

Subconscious: 50–80% lower cost; billed on processed tokens. Inception: $0.20 in, $0.75 out (Mercury 2.5). The cheaper choice depends on the model and workload.

Which has more context, Subconscious or Inception?

Subconscious: 5M+ effective context. Inception: 260K.

Related comparisons

Run your longest agent traces on Subconscious

Point the OpenAI or Anthropic SDK, or the coding agent you already use, at Subconscious. Keep Inception for the work it does best and send the long runs to us.

Get an API keyRead the docs