We raised $5.1M for long-running agents.
Subconscious
vs

Meta vs Thinking Machines

Meta moved to a closed Muse API with a small open model. Thinking Machines went the other way, releasing large open Inkling weights beside its training API.

By The Subconscious Team · Updated

Meta vs Thinking Machines: key differences

Meta Superintelligence Labs serves Muse Spark 1.3 through the Meta Model API, still in public preview, with 1M context at $1.25 in and $4.25 out and endpoints in OpenAI, Anthropic and a stateful agentic format. A Contributor tier drops prices to $0.10 in and $0.20 out if Meta can train on your data. For self-hosting, Meta ships Muse Glimmer, a distilled open model. Thinking Machines' Inkling is priced nearly the same, $1.00 in and $4.05 out with 1M context, but its weights are open under Apache 2.0 at 975B parameters, and it takes text, image and audio input. Its serverless API is also beta.

Customization is where they part. Meta's path is fine-tuning the open Muse Glimmer weights yourself. Thinking Machines' Tinker is a managed trainer: four low-level calls for SFT or RL with LoRA adapters on Inkling, Kimi K2.6, GLM-5.3, Qwen3.5 and more, billed by prefill, sample and train tokens. Meta's API is broader, with Muse Image at $0.01 per image, voice transcription and web search grounding on the same key, and measured speed of about 145 to 233 tokens per second on Muse Spark 1.3. Thinking Machines only serves Inkling. Meta suits cost-sensitive multimodal assistants; Thinking Machines suits teams building their own model.

What Meta and Thinking Machines do

Meta

Meta has moved from open Llama releases toward its own closed API. Meta Superintelligence Labs builds the Muse family, and in July 2026 Meta opened a public preview of the Meta Model API with Muse Spark 1.1, a multimodal reasoning model aimed at agentic coding, tool use and computer use. The current lineup runs through Muse Spark 1.3 with a 1M token context. Standard pricing is $1.25 in and $4.25 out per million tokens, with cached input at $0.15, and the endpoint speaks OpenAI Chat Completions, Anthropic Messages and a stateful agentic format.

Example models: Muse Spark 1.3, Muse Glimmer

Full Meta profile

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

Should you choose Meta or Thinking Machines?

Meta

Choose Meta for

  • Cheap prototyping on the Contributor tier
  • Image generation and transcription on one key
  • Drop-in OpenAI and Anthropic endpoint formats

Thinking Machines

Choose Thinking Machines for

  • Large open weights rather than a closed flagship
  • Managed LoRA training with custom RL loops
  • Audio input on an Apache 2.0 model

Meta vs Thinking Machines at a glance

AttributeMetaThinking Machines
Model accessClosed API; open Muse GlimmerOpen weights
Flagship modelsMuse Spark 1.3, Muse GlimmerInkling, Inkling-Small
Speed~145–233 tok/s on Muse Spark 1.3Unknown
Price$1.25 in, $4.25 out; Contributor tier cheaperPer 1M tokens by prefill, sample, train; Inkling $1.00 in, $4.05 out
CustomizationOpen Muse Glimmer weights to fine-tuneLoRA SFT and RL via Tinker
DeploymentMeta Model API (preview)Training API, beta serverless (Inkling only)
Long context1MInkling up to 1M; Tinker 32K–256K

Frequently asked questions

What is the difference between Meta and Thinking Machines?

Meta moved to a closed Muse API with a small open model. Thinking Machines went the other way, releasing large open Inkling weights beside its training API.

When should I choose Meta over Thinking Machines?

Cheap prototyping on the Contributor tier; Image generation and transcription on one key; Drop-in OpenAI and Anthropic endpoint formats.

When should I choose Thinking Machines over Meta?

Large open weights rather than a closed flagship; Managed LoRA training with custom RL loops; Audio input on an Apache 2.0 model.

Is Meta or Thinking Machines cheaper?

Meta: $1.25 in, $4.25 out; Contributor tier cheaper. Thinking Machines: Per 1M tokens by prefill, sample, train; Inkling $1.00 in, $4.05 out. The cheaper choice depends on the model and workload.

Which has more context, Meta or Thinking Machines?

Meta: 1M. Thinking Machines: Inkling up to 1M; Tinker 32K–256K.

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 MetaSubconscious vs Thinking Machines