Cogito

What it does

Cogito is an LLM inference API from Decart. You send chat completion requests (OpenAI-compatible, including streaming, function calling, and structured JSON outputs) to a Cogito endpoint, and it returns model responses. Its catalog covers open-weight models such as Kimi K2.6, GLM-5.2, GPT-OSS 120B, and Qwen3 235B, with published context windows and throughput figures per model.

Developers use it by changing the base URL and API key in existing OpenAI SDK code, then selecting a model from the catalog. Cogito's distinguishing claim is speed: it runs on a hybrid fleet of AWS Trainium and NVIDIA Blackwell, and its premium ultra-fast tier is advertised at 1,000+ tokens per second on reserved capacity. Standard endpoints are self-serve with per-token pricing.

It suits teams building latency-sensitive or multi-step agents that need open-weight models served fast, and enterprises wanting single-tenant deployments. Access requires an API key from the sign-up page; ultra-fast reserved capacity requires contacting sales. Throughput varies by model and prompt, and some models (such as GPT-OSS 120B) are only available through sales for dedicated capacity.

What it can do

• OpenAI-compatible chat completions API with streaming responses
• Function calling and structured JSON outputs
• Access to open-weight models including Kimi K2.6, GLM-5.2, GPT-OSS 120B, and Qwen3 235B
• Per-model published context windows and throughput figures
• Ultra-fast reserved-capacity tier advertised at 1,000+ tokens per second
• Hard spend caps per organization and per API key

Before you choose

• Throughput figures are benchmarks that vary by model and prompt; the 1,000+ tps figure applies to the ultra-fast reserved tier, not standard endpoints.
• Ultra-fast reserved capacity and some models such as GPT-OSS 120B are sold through sales contact rather than self-serve pricing.
• Standard tier bills per token with cached input priced separately; ultra-fast is billed per workload with spend caps denominated in requests per minute.

Listed under

• Coding
• Sandboxes and infrastructure
• Agents
• API
• Developers
• Fast inference
• LLM inference
• Open source models

Get started

Sign up to get an API key; the docs state you start with free credits and no card is required.

1. Sign up at cogito.decart.ai/sign-up and create an API key.
2. Set the COGITO_API_KEY environment variable and change the OpenAI client base_url to https://api.cogito.decart.ai/v1.
3. Browse the model catalog, choose a model, and send a request using the quickstart snippet.

Standard tier bills prompt and completion tokens per million; cached input is billed at a lower rate when the upstream cache hits. Ultra-fast is priced per workload, not per token. Fine-tuned variants use the same rates. Free credits are mentioned on the homepage ($5, no card) and docs ('free credits'), but the pricing page confirms only that credit does not expire and that $1 of credit equals $1 of usage. Free-tier status is therefore unconfirmed.

Source: cogito.decart.ai/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Cogito | Ultra-fast Inference ↗cogito.decart.ai
• Pricing | Cogito ↗cogito.decart.ai
• Overview - Cogito ↗cogito.decart.ai

Company website

cogito.decart.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.