Grok 4.20 lists a 2,000,000-token context, the longest among the reviewed Grok models, plus tool calling. The same row publishes a knowledge cutoff of 2025-09-01. Use it to read a corpus you supply, not to ask what happened last week. Grok 4.20 is listed at $1.25 input and $2.50 output per million tokens, with a 2,000,000-token context window. In: file, image, text. Out: text. Tools: Yes. Open weights: No public weights id listed.
Tools: Yes.
in file
in image
in text
out text
tools
Facts
Facts for Grok 4.20. OpenRouter models API, checked 20 Nov 2026.
Fact
Value
Source
Release date
31 Mar 2026
OpenRouter field created. This is the listing date the feed publishes. Checked 20 Nov 2026.
Context window
2,000,000 tokens
OpenRouter field context_length. Checked 20 Nov 2026.
Max output
1,800,000 tokens
OpenRouter field top_provider.max_completion_tokens. Checked 20 Nov 2026.
Input price
$1.25 per million tokens. Published per-token rate 0.00000125.
OpenRouter field pricing.prompt, USD per token, multiplied by 1,000,000. Checked 20 Nov 2026.
Output price
$2.50 per million tokens. Published per-token rate 0.0000025.
OpenRouter field pricing.completion, USD per token, multiplied by 1,000,000. Checked 20 Nov 2026.
Batch input price
No batch route listed
Sibling OpenRouter id ending in :batch, field pricing.prompt. Checked 20 Nov 2026.
Batch output price
No batch route listed
Sibling OpenRouter id ending in :batch, field pricing.completion. Checked 20 Nov 2026.
Input modalities
file, image, text
OpenRouter field architecture.input_modalities. Checked 20 Nov 2026.
Output modalities
text
OpenRouter field architecture.output_modalities. Checked 20 Nov 2026.
Tools
Yes
OpenRouter field supported_parameters contains tools. Checked 20 Nov 2026.
Open weights
No public weights id listed
OpenRouter field hugging_face_id. Empty means no public weights id is listed, which is not a legal opinion. Checked 20 Nov 2026.
Knowledge cutoff
2025-09-01
OpenRouter field knowledge_cutoff. Checked 20 Nov 2026.
Reasoning
Supported, optional
OpenRouter field reasoning. Missing means unknown, not unsupported. Checked 20 Nov 2026.
Other OpenRouter pricing fields, shown raw. The documented unit is USD per token, request, or unit — not converted to per million. Checked 20 Nov 2026.
A $0 token price does not mean the model is free to use. Another published price field is non-zero and may be the one you are billed on.
Listing blurb
Grok 4.20 is a reasoning model from SpaceXSI with industry-leading speed and agentic tool calling capabilities. It combines the lowest hallucination rate on the market with strict prompt adherance, delivering...
No third-party index was included for this model in the feed. The Best Model For does not fill that gap.
Guide note · 19 Nov 2026
Grok 4.20 lists a 2,000,000-token context, the longest among the reviewed Grok models, plus tool calling. The same row publishes a knowledge cutoff of 2025-09-01. Use it to read a corpus you supply, not to ask what happened last week.
Put the corpus in the window and ask for a report that quotes it. The cutoff date is a feature: you know which world the model was trained through, and the documents cover the rest.
Grok 4.20 is listed at $1.25 input and $2.50 output per million tokens, with a 2,000,000-token context window.
Tool calls against a huge prompt
The listing highlights agentic tool calling. Pair that with a long input when the tools are simple and the source text is not.
Grok 4.20 is listed at $1.25 input and $2.50 output per million tokens, with a 2,000,000-token context window.
The tasks are editorial illustrations, not measured benchmarks. The prices in them are read from today's feed.
Grok 4.20 is listed at $1.25 input and $2.50 output per million tokens, with a 2,000,000-token context window. No batch route is listed. Other price fields, if any, are on this page and are not converted.
Does Grok 4.20 have open weights?
No public weights id is listed on the OpenRouter row. The Best Model For does not treat that as a legal claim that weights do not exist anywhere.
What is Grok 4.20 good for?
Grok 4.20 lists a 2,000,000-token context, the longest among the reviewed Grok models, plus tool calling. The same row publishes a knowledge cutoff of 2025-09-01. Use it to read a corpus you supply, not to ask what happened last week.