Grok 4.20

Grok 4.20

Quick answer: Grok 4.20 is xAI's high-performance model from March 2026, designed for industry-leading speed and agentic tool calling. It offers a 1M-token context window at $1.25 input / $2.50 output per 1M tokens, includes full reasoning support, and is positioned as a low-cost, high-throughput option between Grok 4 and Grok 4.5. On LHTB it scored 0.080 mean reward (rank 21), solving 0 of 46 tasks.

At a Glance

Where Grok 4.20 leads

• 1M-token context window at $1.25/$2.50 per 1M tokens — one of the best cost-per-context-token ratios among xAI models.
• Industry-leading response speed with agentic tool calling.
• Low hallucination rate with strict prompt adherence.

Where it lags

• LHTB score of 0.080 places it last among 21 models — significantly behind Grok 4.5.
• Preceded by Grok 4.5 as xAI's recommended model for coding and complex tasks.

Best for: High-throughput, latency-sensitive API workloads requiring broad context at low cost.

What Grok 4.20 Is

Grok 4.20 is an intermediate model in xAI's Grok 4.x generation, released March 9, 2026 (model ID: grok-4.20-0309-reasoning). xAI describes it as "a high-performance model with industry-leading speed and agentic tool calling capabilities" that "combines the lowest hallucination rate on the market with strict prompt adherence, delivering consistently precise and truthful responses."

It supports reasoning — the model can think before responding — alongside standard function calling and structured outputs. The 1M-token context window is xAI's largest available, twice the size of Grok 4.5's 500K window, making Grok 4.20 the better choice for extremely long-context use cases where throughput and cost matter more than top-end task performance.

On LHTB — the Long-Horizon Terminal-Bench evaluating 18 frontier models on 46 hard, 90-minute terminal tasks — Grok 4.20 scored 0.080 mean reward, placing last overall. It solved no tasks outright. This result predates the July 2026 evaluation and likely reflects the model's earlier training vintage relative to the Grok 4.5 successor that scored 0.505 at the top of the same benchmark.

Specifications

Related Benchmarks

• Long-Horizon Terminal-Bench (LHTB) — 0.080 mean reward, rank 21 (0/46 solved)