# Command / Command Light

> Cohere's first-generation Command and Command Light text-generation models — a 4K-context instruction follower plus a smaller, faster sibling, both retired in favor of Command R and Command A.

- Maker: Cohere
- Line: Command (legacy)
- Canonical page: https://ai-tldr.dev/models/command-legacy/
- Updated: 2026-06-18

## In short

Command / Command Light is Cohere's first-generation Command and Command Light text-generation models — a 4K-context instruction follower plus a smaller, faster sibling, both retired in favor of Command R and Command A. Cohere released it on 8 November 2022 as an API-only model. It has a 4K-token context window. You can use it through AWS Bedrock and Oracle OCI Generative AI.

## Facts

| Fact | Value | Source |
| --- | --- | --- |
| Context window | 4K tokens (max output 4K tokens) | [1] |
| Input price | $1.50 / 1M tokens (Command); $0.30 / 1M tokens (Command Light) | [2] |
| Output price | $2.00 / 1M tokens (Command); $0.60 / 1M tokens (Command Light) | [2] |
| Licence | Proprietary · API only | [1] |
| Released | 8 November 2022 | [3] |
| Parameters | Command ~52B; Command Light ~6B (per Oracle OCI model listings) | [1] |
| Knowledge cutoff | February 2023 (Command Light, per LLMReference) | [1] |

## Sources

1. [Cohere — Models overview (legacy command specs)](https://docs.cohere.com/docs/models)
2. [AWS Bedrock — pricing (historical)](https://aws.amazon.com/bedrock/pricing/)
3. [Cohere changelog — Major Command Deprecations (2025-09-15)](https://docs.cohere.com/changelog/2025-09-15-major-command-deprecations)

## Overview

Command and Command Light were Cohere's original instruction-following text-generation models, the foundation of the Command family that later grew into Command R, Command R+, and Command A. The lineage traces back to the beta "command-xlarge-20221108" model Cohere introduced in November 2022 — a generative model conditioned to respond well to single-statement commands. The production models were served on AWS Bedrock as cohere.command-text-v14 and cohere.command-light-text-v14, and on Oracle OCI Generative AI as cohere.command (listed at ~52B parameters) and cohere.command-light (~6B parameters).

Command is described by Cohere as "an instruction-following conversational model that performs language tasks with high quality, more reliably and with a longer context than our base generative models." Command Light is "a smaller, faster version of command — almost as capable, but a lot faster," aimed at latency-sensitive and cost-sensitive workloads. Both models share a 4,000-token (4K) context window and a 4,000-token maximum output, and both are decoder-only Transformers focused on English text generation, summarization, and classification rather than the retrieval-augmented generation and tool use that define the later Command R series.

These are legacy models. On September 15, 2025 Cohere deprecated both command and command-light (alongside command-r-03-2024 and command-r-plus-04-2024), and directs users to command-r-08-2024, command-r-plus-08-2024, or command-a-03-2025 — the last being its strongest current model. Oracle has retired the on-demand cohere.command and cohere.command-light models on OCI Generative AI, and fine-tuning for these models has been wound down. New projects should build on the current Command R or Command A models; this page documents the original generation for historical reference and migration.

## Pricing

- Input: $1.50 / 1M tokens (Command); $0.30 / 1M tokens (Command Light) per 1M tokens
- Output: $2.00 / 1M tokens (Command); $0.60 / 1M tokens (Command Light) per 1M tokens

AWS Bedrock on-demand pricing while the models were active (Command: $0.0015/1K input, $0.0020/1K output; Command Light: $0.0003/1K input, $0.0006/1K output). Models are now deprecated/retired; pricing is historical.

Source: https://aws.amazon.com/bedrock/pricing/

## Strengths

- Simple, reliable instruction following for single-statement commands and short conversational prompts
- Command Light offered notably faster, cheaper generation when full Command quality was not required
- Straightforward text generation, summarization, and classification without the complexity of tool-use or RAG orchestration
- Available across multiple clouds during its lifetime (Cohere API, AWS Bedrock, Oracle OCI Generative AI)

## Best for

- Short-form copy, drafting, and conversational responses (pre-RAG generation)
- Text summarization and rewriting via the legacy generate/summarize endpoints
- Lightweight classification and content tasks where Command Light's speed and low cost mattered
- Migration reference for teams moving older command / command-light integrations to Command R or Command A

## How to access

- [AWS Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-cohere-command.html) — model id `cohere.command-text-v14`
- [AWS Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-cohere-command.html) — model id `cohere.command-light-text-v14`
- [Oracle OCI Generative AI](https://docs.oracle.com/en-us/iaas/Content/generative-ai/cohere-command.htm) — model id `cohere.command (retired)`
- [Oracle OCI Generative AI](https://docs.oracle.com/en-us/iaas/Content/generative-ai/cohere-command-light.htm) — model id `cohere.command-light (retired)`

## Command (legacy) versions

- Command / Command Light — this page

## FAQ

### Are Command and Command Light still available?

No. Cohere deprecated both command and command-light on September 15, 2025, and they are no longer offered for on-demand use on Oracle OCI Generative AI. Cohere recommends migrating to command-r-08-2024, command-r-plus-08-2024, or command-a-03-2025.

### What is the context window of the legacy Command models?

Both Command and Command Light have a 4,000-token (4K) context window and a 4,000-token maximum output, per Cohere's model documentation. This is far smaller than the later Command R (128K) and Command A (256K) models.

### How big were Command and Command Light?

Oracle's OCI Generative AI listings describe the original Command as a ~52B-parameter model and Command Light as ~6B parameters. Cohere did not publish a detailed parameter breakdown of its own; these figures come from the OCI model pages.

### What should I use instead of Command or Command Light?

For most workloads, Cohere recommends Command A (command-a-03-2025), its strongest current model, or the Command R series (command-r-08-2024 and command-r-plus-08-2024), which add long context, retrieval-augmented generation, and tool use that the legacy Command models lack.

## Links

- [Cohere docs — Command & Command Light](https://docs.cohere.com/v1/docs/command-beta)
- [Cohere — Models overview (legacy command specs)](https://docs.cohere.com/docs/models)
- [Cohere — Deprecations](https://docs.cohere.com/docs/deprecations)
- [Cohere changelog — Major Command Deprecations (2025-09-15)](https://docs.cohere.com/changelog/2025-09-15-major-command-deprecations)
- [Cohere changelog — Command beta introduced (command-xlarge-20221108)](https://docs.cohere.com/changelog/improvements-to-current-models-new-beta-model-command)
- [AWS Bedrock — Cohere Command model parameters](https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-cohere-command.html)
- [AWS Bedrock — pricing (historical)](https://aws.amazon.com/bedrock/pricing/)
- [Oracle OCI — Cohere Command (52B), retired](https://docs.oracle.com/en-us/iaas/Content/generative-ai/cohere-command.htm)
- [Oracle OCI — Cohere Command Light, retired](https://docs.oracle.com/en-us/iaas/Content/generative-ai/cohere-command-light.htm)

Source: AI/TLDR LLM registry, https://ai-tldr.dev/models/command-legacy/