Grok 4.20
xAI released 2026-03-31 proprietary
No independent quality score.
No benchmark we track has measured this model. That is not the same as measuring it and finding it wanting — we simply cannot rank it, so we do not.
Our take
editorial — not a measurementThe context-window outlier: 2M tokens at $1.25/$2.50, which is the cheapest long-context option from any major proprietary vendor. Same 200k tier threshold as the rest of the Grok line, doubling to $2.50/$5 above it — so the 2M window is only cheap in headline terms, and a genuinely 2M-token prompt bills entirely at the upper tier. Still, for bulk long-document processing the arithmetic is hard to beat. A multi-agent variant exists at the same rate.
Strengths
- 2M token context window — largest here
- $1.25/$2.50 short-context, very cheap
- Multi-agent variant at identical pricing
- Published tier threshold
Weaknesses
- Everything past 200k bills at $2.50/$5, so the big window is not cheap in practice
- Below the 4.5/4.6 line on capability
- No batch discount published
Reach for it when
- Bulk long-document ingestion
- Cheap very-long-context retrieval
Avoid it if
- You need frontier reasoning quality
- You assumed the headline rate applies at 2M tokens
Every price dimension
| Input | $1.25/M |
|---|---|
| Output | $2.50/M |
| Cached input | $0.200/M |
| Web search | $0.005/call |
Past 200,000 tokens the price changes. Input goes to $2.50/M (2×) and output to $5.00/M. The headline rate does not apply to a long-context workload.
Independent scores
| LMArena Elo | 1444.4beta1 |
|---|
Sources are listed separately rather than averaged. Across the 62 models both have scored they correlate at r = 0.806 — close agreement overall, but four models rank very differently between them, and a blended score would hide exactly those.
Best rankings by task
- htmlslides #12of 34 1180
- asciiart #20of 83 1205
- webapps #23of 56 1181
- godotgamedev #26of 43 1089
- svg #30of 98 1203
- fullstack #30of 59 1086
Capability
| Context window | 2M |
|---|---|
| Input modes | text, image, file |
| Tool use | yes |
| Reasoning | optional |
| Knowledge cutoff | 2025-09-01 |
| Open weights | no |
Provenance
| Price source | openrouter.ai |
|---|---|
| Fetched | 2026-08-24 |
| Quality data | unrated |
| Cross-checked | vendor page |
Grok 4.20 is a reasoning model from SpaceXAI with industry-leading speed and agentic tool calling capabilities. It combines the lowest hallucination rate on the market with strict prompt adherance, delivering...