Kimi K3 (batch)

Overview

Kimi K3 (batch) is an open-weight multimodal reasoning model from MoonshotAI. The model has 2.8 trillion parameters and supports text, image, and video inputs, with text output. Its capabilities include reasoning and tool use. MoonshotAI lists it for complex coding, knowledge work, and long-running agent workflows.

The batch model has a context window of 1,048,576 tokens and a maximum output of 16,384 tokens. Those limits make context capacity a notable part of its profile, while the input modalities allow it to work with more than text alone. The listed model ID is moonshotai/kimi-k3:batch.

Kimi K3 belongs to MoonshotAI’s model family and is identified as open-weight. Its Hugging Face ID is moonshotai/Kimi-K3. It was released on 2026-07-16.

Key features

• Multimodal inputs: Accepts text, images, and video, and produces text.
• Reasoning: Reasoning is supported, alongside tool use.
• Large context: The context window is 1,048,576 tokens.
• Output limit: Up to 16,384 tokens per response.
• Open weights: The model is listed as open-weight, with a Hugging Face identifier provided.
• Model scale: Kimi K3 has 2.8 trillion parameters.

MoonshotAI positions Kimi K3 for demanding coding and knowledge tasks as well as agentic workflows that extend over longer horizons. The available description does not specify benchmark scores or more detailed task-level results, so those claims are best read as intended use rather than a quantified performance ranking.

Pricing

The listed batch API rates are USD 2.28 per 1 million input tokens and USD 11.4 per 1 million output tokens. Output tokens cost five times the input rate, so workloads that generate substantial responses will have a materially different token bill from input-heavy tasks. The provided pricing is per million tokens; no other billing terms are specified here.

Platforms

Kimi K3 (batch) is identified by the model ID moonshotai/kimi-k3:batch. Its stated inputs are text, image, and video, with text as the output modality. The model entry also lists tool support and a 1,048,576-token context window.

Who it's for

Kimi K3 may suit teams evaluating an open-weight model for coding, knowledge work, or extended agent workflows, particularly where image or video input and a large context window matter. Its batch pricing can also be relevant to token-budget planning, but the output rate should be weighed against expected response volume.

It is a less straightforward fit for buyers who need evidence of measured performance before choosing a model: the available details provide intended use and technical limits, but no benchmarks. Teams should also assess how its stated modality, tools, context, and token rates map to their own integration and workload requirements.

Pros and cons

Pros

• Open-weight model with a stated Hugging Face ID.
• Accepts text, image, and video inputs.
• Supports reasoning and tool use.
• Offers a 1,048,576-token context window.
• Designed for coding, knowledge work, and longer agentic workflows.

Cons

• The output price is USD 11.4 per 1 million tokens, substantially above the USD 2.28 input rate.
• The supplied details do not include benchmark results or measured comparisons.
• Text is the only listed output modality.

Alternatives

For other models in the same family, consider Kimi K3, Kimi K2.7 Code, Kimi Latest, Kimi K2.6, Kimi K2.5, Kimi K2 Thinking, Kimi K2 0905, and Kimi K2 0711.

Browse the MoonshotAI models list for more from the lab. Other useful directories include open-weight models, vision language models, reasoning models, longest-context models, and newest models.

Verdict

Kimi K3 (batch) combines open weights, reasoning and tool support, multimodal input, and a very large stated context window in a model aimed at complex coding, knowledge work, and long-running agent workflows. Its rates are clear: USD 2.28 per 1 million input tokens and USD 11.4 per 1 million output tokens. The key trade-off for teams is whether the model’s context and input capabilities suit their workload enough to justify the higher output-token cost. With no benchmarks supplied, its intended use cases should not be mistaken for verified comparative performance.