Grok 4.20 vs Grok 4.3
Updated October 21, 2026. 46 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick Grok 4.3 for coding: it scores 37.4 on the Vibe Coding Index against 33.3 for Grok 4.20.
Pick Grok 4.20 when the context window is larger, 2M against 1M tokens, or when what matters is SWE-bench Verified (Vals AI) or Terminal-Bench 2.1 (Vals AI), where it leads.
- Shared results
- 46
- Grok 4.20 only
- 8
- Grok 4.3 only
- 58
- Like-for-like categories
- 7 of 9
Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
Grok 4.3
Grok 4.3 scores 37.4 against 33.3.
2 and 3 benchmarks behind the scores
Intelligence
AA Intelligence Index
Grok 4.20
Grok 4.20 leads, 25.7 against 24.9.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Grok 4.3 has a AA Agentic Index result, 15.5.
Directional only
Reasoning
GPQA Diamond
Grok 4.3
Grok 4.3 leads, 91.4% against 88.6%.
Same board, both measured
Math
AIME
No call
Only Grok 4.20 has a AIME result, 96.5%.
Directional only
Knowledge
MMLU Pro
Grok 4.20
Grok 4.20 leads, 86.3% against 85.8%.
Same board, both measured
Web design
Design Arena: websites
Grok 4.20
Grok 4.20 leads, 1229 against 1196.
Same board, both measured
Long documents
Listed context window
Grok 4.20
Grok 4.20 takes 2M tokens against 1M.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
No call
Grok 4.20 $0.0025, Grok 4.3 $0.0025 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
Grok 4.3
Grok 4.20 $0.72, Grok 4.3 $0.31.
Published attempt cost and pass rate
At a glance
- Coding index
- Grok 4.20 33.3 from 2 benchmarks, Grok 4.3 37.4 from 3 benchmarks.
- Widest category
- Agentic work, 9.8 points apart on average over 12 shared benchmarks.
- Widest single gap
- Vending-Bench 2: Grok 4.20 $4,663 against Grok 4.3 $35, 28.9 points apart on a 0 to 100 scale.
- Output price
- Both cost $2.50 per million output tokens.
- Context window
- Grok 4.20 takes 2M tokens, Grok 4.3 1M.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Grok 4.20 | Bars | Grok 4.3 | Ties |
|---|---|---|---|---|
| Coding | 3 | 3 | 1 tie | |
| Intelligence | 1 | 0 | ||
| Agentic work | 8 | 4 | ||
| Reasoning | 2 | 2 | ||
| Knowledge | 2 | 5 | ||
| Web design | 8 | 0 | ||
| General | 2 | 5 | ||
| All shared boards | 26 | 19 | 1 tie |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index33
| Benchmark | Grok 4.20 | Grok 4.3 | Leader |
|---|---|---|---|
| LMArena WebDev | Not measured | 1358 | Only Grok 4.3 measured |
| AA Coding Index | Not measured | 42.2 | Only Grok 4.3 measured |
| LiveCodeBench (Vals AI) | 84.3% | 84.5% | Grok 4.3 |
| SWE-bench Verified (Vals AI) | 72.2% | 71.4% | Grok 4.20 |
| Vibe Code Bench (Vals AI) | 4.1% | 19.4% | Grok 4.3 |
| Code migration (Vals AI) | 0.3% | 6.8% | Grok 4.3 |
| ProgramBench (Vals AI) | Not measured | 0.0% | Only Grok 4.3 measured |
| Terminal-Bench 2.1 (Vals AI) | 44.2% | 42.0% | Grok 4.20 |
| SkillsBench (Vals AI) | Not measured | 40.6% | Only Grok 4.3 measured |
| SciCode | Not measured | 48.3% | Only Grok 4.3 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 37.9% | 37.9% | Tie |
| Terminal-Bench | 57.3% | Not measured | Only Grok 4.20 measured |
| WeirdML | 52.3% | 49.9% | Grok 4.20 |
| ALE-Bench | Not measured | 944.2 | Only Grok 4.3 measured |
| LMArena Text: coding | Not measured | 1415 | Only Grok 4.3 measured |
| LMArena Text: software and IT | Not measured | 1416 | Only Grok 4.3 measured |
| LMArena WebDev: React | Not measured | 1353 | Only Grok 4.3 measured |
| LMArena WebDev: HTML | Not measured | 1367 | Only Grok 4.3 measured |
| LMArena WebDev: from an image | Not measured | 1375 | Only Grok 4.3 measured |
| Neither model has a result on the other 19 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Grok 4.20 | Grok 4.3 | Leader |
|---|---|---|---|
| AA Intelligence Index | 25.7 | 24.9 | Grok 4.20 |
| Epoch Capabilities Index | 152.1 | Not measured | Only Grok 4.20 measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index84
| Benchmark | Grok 4.20 | Grok 4.3 | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 15.5 | Only Grok 4.3 measured |
| Design Arena: agents building web apps | 1144 | 1126 | Grok 4.20 |
| Design Arena: agents building full-stack apps | 1054 | 1006 | Grok 4.20 |
| Design Arena: agents building mobile apps | 1090 | 1059 | Grok 4.20 |
| Design Arena: agents building Android apps | 1052 | 968 | Grok 4.20 |
| Design Arena: agents building Godot games | 1089 | 914 | Grok 4.20 |
| Design Arena: agents building games | Not measured | 1007 | Only Grok 4.3 measured |
| Design Arena: HTML slides | 1180 | 1008 | Grok 4.20 |
| Design Arena: PowerPoint slides | Not measured | 1013 | Only Grok 4.3 measured |
| Finance Agent (Vals AI) | 28.5% | 37.7% | Grok 4.3 |
| Tax Agent Bench (Vals AI) | 7.0% | 6.2% | Grok 4.20 |
| Legal Research Bench (Vals AI) | 13.9% | 15.4% | Grok 4.3 |
| Excel Modeling Benchmark (Vals AI) | 11.7% | 18.4% | Grok 4.3 |
| τ²-Bench Telecom | 93.0% | 97.7% | Grok 4.3 |
| Vending-Bench 2 | $4,663 | $35 | Grok 4.20 |
| LMArena Search | Not measured | 1166 | Only Grok 4.3 measured |
| Neither model has a result on the other 23 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index22
| Benchmark | Grok 4.20 | Grok 4.3 | Leader |
|---|---|---|---|
| GPQA Diamond | 88.6% | 91.4% | Grok 4.3 |
| MMMU Pro (Vals AI) | 83.5% | 83.1% | Grok 4.20 |
| GPQA Diamond (Artificial Analysis run) | 91.1% | 90.1% | Grok 4.20 |
| Humanity's Last Exam (Artificial Analysis run) | 34.5% | 37.2% | Grok 4.3 |
| GPQA Diamond (Epoch AI run) | Not measured | 88.8% | Only Grok 4.3 measured |
| ARC-AGI | 89.5% | Not measured | Only Grok 4.20 measured |
| ARC-AGI-2 | 65.1% | Not measured | Only Grok 4.20 measured |
| Chess puzzles | Not measured | 18.0% | Only Grok 4.3 measured |
| LMArena Text: hard prompts | Not measured | 1396 | Only Grok 4.3 measured |
| LMArena Text: expert | Not measured | 1384 | Only Grok 4.3 measured |
| LMArena Text: hard prompts in English | Not measured | 1402 | Only Grok 4.3 measured |
| LMArena Vision: diagrams | Not measured | 1241 | Only Grok 4.3 measured |
| LMArena Vision: homework | Not measured | 1241 | Only Grok 4.3 measured |
| Neither model has a result on the other 8 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index7 boards, none shared
| Benchmark | Grok 4.20 | Grok 4.3 | Leader |
|---|---|---|---|
| AIME | 96.5% | Not measured | Only Grok 4.20 measured |
| OTIS Mock AIME | Not measured | 93.3% | Only Grok 4.3 measured |
| FrontierMath, tiers 1 to 3 | Not measured | 42.8% | Only Grok 4.3 measured |
| FrontierMath, tier 4 | Not measured | 14.6% | Only Grok 4.3 measured |
| ProofBench | Not measured | 11.0% | Only Grok 4.3 measured |
| LMArena Text: math | Not measured | 1387 | Only Grok 4.3 measured |
| LMArena Text: mathematical work | Not measured | 1396 | Only Grok 4.3 measured |
| Neither model has a result on the other 6 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index25
| Benchmark | Grok 4.20 | Grok 4.3 | Leader |
|---|---|---|---|
| MMLU Pro | 86.3% | 85.8% | Grok 4.20 |
| LegalBench (Vals AI) | 77.7% | 84.5% | Grok 4.3 |
| CaseLaw (Vals AI) | 54.4% | 79.3% | Grok 4.3 |
| CorpFin (Vals AI) | 63.7% | 68.5% | Grok 4.3 |
| TaxEval (Vals AI) | 74.1% | 70.8% | Grok 4.20 |
| MortgageTax (Vals AI) | 45.4% | 48.3% | Grok 4.3 |
| MedQA (Vals AI) | 94.5% | Not measured | Only Grok 4.20 measured |
| MedCode (Vals AI) | 32.2% | 38.1% | Grok 4.3 |
| SimpleQA Verified | Not measured | 33.2% | Only Grok 4.3 measured |
| LMArena Text: science | Not measured | 1399 | Only Grok 4.3 measured |
| LMArena Text: medicine and healthcare | Not measured | 1384 | Only Grok 4.3 measured |
| LMArena Text: legal and government | Not measured | 1386 | Only Grok 4.3 measured |
| LMArena Text: business and finance | Not measured | 1381 | Only Grok 4.3 measured |
| Neither model has a result on the other 4 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index80
| Benchmark | Grok 4.20 | Grok 4.3 | Leader |
|---|---|---|---|
| Design Arena: websites | 1229 | 1196 | Grok 4.20 |
| Design Arena: UI components | 1199 | 1192 | Grok 4.20 |
| Design Arena: game dev | 1192 | 1174 | Grok 4.20 |
| Design Arena: data viz | 1217 | 1180 | Grok 4.20 |
| Design Arena: 3D | 1193 | 1125 | Grok 4.20 |
| Design Arena: all code categories | 1221 | 1187 | Grok 4.20 |
| Design Arena: SVG | 1164 | 1082 | Grok 4.20 |
| Design Arena: ASCII art | 1174 | 1125 | Grok 4.20 |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index25
| Benchmark | Grok 4.20 | Grok 4.3 | Leader |
|---|---|---|---|
| Public Benefits Bench (Vals AI) | Not measured | 51.7% | Only Grok 4.3 measured |
| SAGE (Vals AI) | 38.2% | 19.7% | Grok 4.20 |
| MedScribe (Vals AI) | 63.4% | 74.4% | Grok 4.3 |
| Vals Multimodal Index | 39.1% | 43.3% | Grok 4.3 |
| IFBench | 81.2% | 83.3% | Grok 4.3 |
| Long context reasoning (AA-LCR) | 69.0% | 75.0% | Grok 4.3 |
| DTBench | 90.1% | 90.7% | Grok 4.3 |
| LMCA | 38.7% | 38.3% | Grok 4.20 |
| CL-bench | 22.2% | Not measured | Only Grok 4.20 measured |
| CL-bench Life | 11.9% | Not measured | Only Grok 4.20 measured |
| LMArena Text | Not measured | 1397 | Only Grok 4.3 measured |
| LMArena Text: instruction following | Not measured | 1366 | Only Grok 4.3 measured |
| LMArena Text: creative writing | Not measured | 1380 | Only Grok 4.3 measured |
| LMArena Vision | Not measured | 1229 | Only Grok 4.3 measured |
| LMArena Text: English | Not measured | 1404 | Only Grok 4.3 measured |
| LMArena Text: non-English | Not measured | 1385 | Only Grok 4.3 measured |
| LMArena Text: Chinese | Not measured | 1422 | Only Grok 4.3 measured |
| LMArena Text: German | Not measured | 1395 | Only Grok 4.3 measured |
| LMArena Text: Japanese | Not measured | 1379 | Only Grok 4.3 measured |
| LMArena Text: Korean | Not measured | 1354 | Only Grok 4.3 measured |
| LMArena Text: French | Not measured | 1411 | Only Grok 4.3 measured |
| LMArena Text: Spanish | Not measured | 1399 | Only Grok 4.3 measured |
| LMArena Text: Russian | Not measured | 1399 | Only Grok 4.3 measured |
| LMArena Text: Polish | Not measured | 1398 | Only Grok 4.3 measured |
| LMArena Text: multi-turn | Not measured | 1406 | Only Grok 4.3 measured |
| LMArena Text: longer queries | Not measured | 1393 | Only Grok 4.3 measured |
| LMArena Text: writing and language | Not measured | 1385 | Only Grok 4.3 measured |
| LMArena Text: entertainment and media | Not measured | 1370 | Only Grok 4.3 measured |
| LMArena Vision: English | Not measured | 1235 | Only Grok 4.3 measured |
| LMArena Vision: Chinese | Not measured | 1257 | Only Grok 4.3 measured |
| LMArena Vision: reading text | Not measured | 1235 | Only Grok 4.3 measured |
| LMArena Vision: humor | Not measured | 1222 | Only Grok 4.3 measured |
| LMArena Vision: recognizing things | Not measured | 1217 | Only Grok 4.3 measured |
| LMArena Vision: creative writing | Not measured | 1237 | Only Grok 4.3 measured |
| Neither model has a result on the other 12 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Grok 4.20 | Grok 4.3 | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Same cost
Repository review
50K input + 3K output tokens
Same cost
Agent loop
200K input + 10K output tokens
Same cost
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | SpaceXAI Grok 4.20 | SpaceXAI Grok 4.3 |
|---|---|---|
| Maker | xAI | xAI |
| Weights | No public weights listed | No public weights listed |
| Released | March 31, 2026 | April 30, 2026 |
| Context window | 2M | 1M |
| Input $/Mtok | $1.25 | $1.25 |
| Output $/Mtok | $2.50 | $2.50 |
| Coding index | 33.3 | 37.4 |
| Independent measurements behind it | 2 | 3 |
| Cost per solved task | $0.72 | $0.31 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Grok 4.20 better than Grok 4.3 for coding?
Pick Grok 4.3 for coding: it scores 37.4 on the Vibe Coding Index against 33.3 for Grok 4.20. Pick Grok 4.20 when the context window is larger, 2M against 1M tokens, or when what matters is SWE-bench Verified (Vals AI) or Terminal-Bench 2.1 (Vals AI), where it leads.
Which is cheaper, Grok 4.20 or Grok 4.3?
Grok 4.20 costs $1.25 per million input tokens and $2.50 per million output tokens; Grok 4.3 costs $1.25 and $2.50. Per solved SWE-bench Verified task, Grok 4.20 costs $0.72 and Grok 4.3 $0.31.
Which has the bigger context window, Grok 4.20 or Grok 4.3?
Grok 4.20, with 2M tokens against 1M.
How many benchmarks is the Grok 4.20 vs Grok 4.3 comparison based on?
46 benchmarks have a result for both models as of October 21, 2026. Grok 4.20's coding index rests on 2 independent benchmarks and Grok 4.3's on 3. Where only one model has a result, the page says so instead of naming a winner.