Grok 4.20

Overview

Grok 4.20 is a reasoning model from SpaceXAI, identified as x-ai/grok-4.20. It accepts text, images and files, and produces text. The listed capabilities include vision and tool use, placing it in scope for workflows that combine language tasks with visual inputs or tool-mediated actions.

Its 2,000,000-token context window is a notable fit for applications that need to supply unusually large bodies of input in a single context. The listed maximum output is 1,800,000 tokens. These are separate limits: the maximum output does not by itself describe how much input and output can be accommodated together in one request.

The model is marked as a reasoning model and as supporting tools. The summary also characterizes it as fast, capable of agentic tool calling, and strict in following prompts, while claiming a low hallucination rate. Those are supplied descriptions, not quantified benchmark results, so teams should treat them as positioning rather than a measured guarantee.

Grok 4.20 was released on March 31, 2026. It belongs to the x-ai family and is provided by SpaceXAI.

Key features

Large context and long outputs

A 2-million-token context window gives builders room to pass extensive material into a request, such as long documents or large collections of text. The 1.8-million-token maximum output is also unusually large as a published limit. Applications still need to account for their actual combined request constraints and the cost of the tokens they send and receive.

Text, images and files

Grok 4.20 accepts text, image and file inputs, with vision listed as a capability. This makes it relevant to systems that need a model to work across textual and visual material, rather than text alone. The available facts do not specify supported file formats or image-processing limits.

Reasoning and tools

Reasoning and tool support are both listed. The model summary describes agentic tool calling and prompt adherence, suggesting a potential role in workflows where a model selects or invokes tools as part of a task. No particular tools, integrations, or execution behavior are specified here, so implementation fit depends on the interface and tool setup available to a team.

Explore Reasoning models, Longest-context models, Vision language models, Models with tool calling, SpaceXAI models and Newest models.

Pricing

Listed pricing is $1.25 USD per 1M input tokens and $2.5 USD per 1M output tokens. Input and output are billed at different rates, so teams should estimate both sides of their workload rather than using a single blended token assumption. Large context and output limits provide headroom, but do not make extensive usage inexpensive by default; actual spend depends on token volume.

Platforms

The supplied model record identifies the model as x-ai/grok-4.20 and gives an OpenRouter page as its official site. It accepts text, images and files and returns text. No additional platform, SDK, deployment, or API compatibility details are specified, so teams should confirm those requirements against their intended stack before adopting it.

Who it's for

Grok 4.20 is worth considering for builders who need a large-context reasoning model with text, image and file inputs, or who are evaluating tool-enabled workflows. Its listed token limits may be useful for applications that pass substantial material or need long generated responses. Teams comparing options should weigh the stated per-token rates against expected input and output volume and validate the qualitative claims against their own requirements.

Pros and cons

Pros

• Large 2,000,000-token context window and 1,800,000-token maximum output.
• Text, image and file inputs, with vision listed.
• Reasoning and tool support are listed.
• Input pricing is $1.25 USD per 1M tokens, and output pricing is $2.5 USD per 1M tokens.

Cons

• The supplied facts do not quantify speed, hallucination rate, or prompt adherence, despite the summary's favorable characterizations.
• Specific supported tools, file formats, deployment choices and API integration details are not provided.
• Large token limits can enable high usage, while input and output tokens incur separate charges.

Alternatives

For a related option in the same model family, consider Grok 4.20 Multi-Agent. Other listed alternatives include GPT-5.4, GPT-5.4 (batch), GPT-5.4 Pro, GPT-5.4 Pro (batch), GPT-5.5, GPT-5.5 (batch) and GPT-5.5 Pro. The available facts do not provide comparable feature or price details for these models, so a direct trade-off cannot be established here.

Verdict

Grok 4.20's clearest strengths on paper are its exceptionally large context and output limits, multimodal inputs, and listed reasoning and tool capabilities. Its published token pricing is specific, but the available information leaves key integration details and performance claims unquantified. It is a candidate for teams that need substantial context or mixed text-and-image inputs; the decision should hinge on how its actual API behavior, tool setup and token costs fit the workload.