Grok 4.20 (Reasoning)
| Released | 2026-03-09 |
|---|---|
| Price per 1M | $1.25 in · $2.50 out |
| Overall score | 53.4 · rank 87 |
| Context window | 1,000,000 tokens |
| Provider | xAI |
| Weights | Proprietary |
Scores
| Dimension | Score | Rank | Benchmarks |
|---|---|---|---|
| Overall: Balanced | 53.4 | 87 | 22 |
| Overall: Coding | 51.8 | 95 | 13 |
| Overall: Reasoning | 56.3 | 65 | 13 |
| Overall: Agentic work | 54.1 | 82 | 12 |
| Overall: Frontend | 43.9 | 149 | 4 |
| Coding | 54.2 | 83 | 2 |
| Agents | 52.5 | 91 | 4 |
| Reasoning | 60.0 | 46 | 6 |
| Math | 55.2 | 68 | 4 |
| Knowledge | 49.9 | 92 | 3 |
| Frontend | 36.8 | 118 | 1 |
| General | 52.3 | 85 | 1 |
| Multimodal | 55.5 | 45 | 1 |
| Value vs price | +4.3 | 82 |
Configurations
| Configuration | Balanced | Benchmarks |
|---|---|---|
| Unspecified setting | 53.7 ±1.3 | 21 |
| High | 51.9 ±6.1 | 1 |
Evidence
Coding 54.2 ±3.3
| Benchmark | Result | Source | Date |
|---|---|---|---|
| SWE-bench Verified (Vals run) | 72.2% | Vals AI | 2026-09-01 |
| LiveCodeBench (Vals run) | 84.3% | Vals AI | 2026-09-01 |
Agents 52.5 ±2.0
| Benchmark | Result | Source | Date |
|---|---|---|---|
| tau2-bench Banking Knowledge | 18.0% | tau2-bench | 2026-05-05 |
| Tax Agent Bench | 7.0% | Vals AI | 2026-11-09 |
| Terminal-Bench 2.0 | 57.3% | Epoch AI | 2026-03-10 |
| Terminal-Bench 2.1 (Vals run) | 44.2% | Vals AI | 2026-10-01 |
| Vending-Bench 2 | $4,663 | Epoch AI | 2026-11-11 |
Reasoning 60.0 ±1.7
| Benchmark | Result | Source | Date |
|---|---|---|---|
| Extended NYT Connections | 85.4% | Lech Mazur | 2026-11-10 |
| ARC-AGI-2 | 65.1% | Epoch AI | 2026-11-11 |
| ARC-AGI-2 (ARC Prize verified) | 65.1% | ARC Prize | 2026-11-11 |
| ARC-AGI-1 (ARC Prize verified) | 89.5% | ARC Prize | 2026-11-11 |
| ARC-AGI-1 | 89.5% | Epoch AI | 2026-11-11 |
| GPQA Diamond | 89.3% | Epoch AI | 2026-07-13 |
| GPQA Diamond (Vals run) | 88.6% | Vals AI | 2026-09-01 |
| Kagi Reasoning Benchmark | 75.0% | Kagi | 2026-11-11 |
| Chess Puzzles | 29.0% | Epoch AI | 2026-10-13 |
Math 55.2 ±1.3
| Benchmark | Result | Source | Date |
|---|---|---|---|
| FrontierMath Tiers 1-3 | 44.9% | Epoch AI | 2026-07-13 |
| FrontierMath Tier 4 | 17.1% | Epoch AI | 2026-07-13 |
| ProofBench | 14.0% | Epoch AI | 2026-11-11 |
| OTIS Mock AIME 2024-2025 | 92.2% | Epoch AI | 2026-07-13 |
Knowledge 49.9 ±3.6
| Benchmark | Result | Source | Date |
|---|---|---|---|
| SimpleQA Verified | 30.2% | Epoch AI | 2026-08-27 |
| MMLU-Pro (Vals run) | 86.3% | Vals AI | 2026-09-01 |
| Legal Research Bench | 13.9% | Vals AI | 2026-11-09 |
Frontend 36.8 ±0.8
| Benchmark | Result | Source | Date |
|---|---|---|---|
| Vibe Code Bench v1.1 | 4.1% | Vals AI | 2026-11-09 |
General 52.3 ±3.3
| Benchmark | Result | Source | Date |
|---|---|---|---|
| LLM Debate Benchmark | 1,404 | Lech Mazur | 2026-10-26 |
Multimodal 55.5 ±2.7
| Benchmark | Result | Source | Date |
|---|---|---|---|
| MMMU-Pro (Vals run) | 83.5% | Vals AI | 2026-09-01 |
Results not used in the score
| Benchmark | Result | Source | Why not used |
|---|---|---|---|
| Terminal-Bench 2.0 | 57.3% | Epoch AI | A stronger configuration result was used |
Names matched to this model
grok-4-2, grok-4-20, grok-4-20-0309-reasoning, grok-4-20-reasoning, grok-4-20-thinking