SpaceXAIGrok 4.20
Grok 4.20 is a SpaceXAI proprietary reasoning model, released 17 Feb 2026. Its best configuration (thinking reasoning effort) ranks #148 of 785 on the BenchLeader Index at 58.6 ±3.8, in the upper half. The ± is the point: 231 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (79) and lowest in agents & tools (50). At $1.56 per million tokens blended it is pricier than most ranked models. Output speed of 112 tokens per second puts it faster than most, with a first answer in 17.1 s. It has been measured at 3 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 22 Oct 2026.
- Blended price
- $1.56/M
- $1.25 in · $2.50 out
- Output speed
- 112 tok/s
- measured by Artificial Analysis
- First answer
- 17 s
- first token 0.70 s
- Context
- 2M
- Full answer
- 22 s
- median, reasoning included
- Cost per run
- –
- Released
- 17 Feb 2026
- Overall index59
- Reasoning59
- Coding54
- Agents & tools50
- Maths54
- Knowledge56
- Instruction following79
- Human preference66
- Multimodal59
- Long context59
- Composite58
Versions
SpaceXAI has shipped 11 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| Grok 4.7high | 21 Sept 2026 | 63.9 | #59 |
| Grok 4.6medium | 11 Aug 2026 | 64.2 | #56 |
| Grok 4.5high | 8 Jul 2026 | 59.3 | #134 |
| Grok 4.3medium | 17 Apr 2026 | 58.7 | #147 |
| Grok 4.2high | 10 Mar 2026 | – | – |
| Grok 4.20thinkingthis page | 17 Feb 2026 | 58.6 | #148 |
| Grok 4.1 | 17 Nov 2025 | 57.2 | #184 |
| Grok 4 | 9 Jul 2025 | 58.0 | #161 |
| Grok 3 | 17 Feb 2025 | 51.5 | #340 |
| Grok 2 | 12 Dec 2024 | 41.9 | #612 |
| Grok 4.20 1 | – | – | – |
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | 46.0 | #502 | 102 tok/s | 0.52 s | $0.0013 | Agents & tools 48 · Composite 45 · Instruction following 51 · Knowledge 41 · Long context 36 · Multimodal 48 · Reasoning 41 |
| thinkingbest | 58.6 | #148 | 112 tok/s | 17 s | $0.0013 | Agents & tools 50 · Coding 54 · Composite 58 · Human preference 66 · Instruction following 79 · Knowledge 56 · Long context 59 · Maths 54 · Multimodal 59 · Reasoning 59 |
| not stated | – | – | 82 tok/s | 0.70 s | $0.0013 | Agents & tools 51 · Coding 51 · Reasoning 42 |
How its index has moved
9 Sept 2026 to 23 Oct 2026- thinking
- no reasoning
The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| GPQA Diamond | – | 89.3%#54 | – | Epoch AI Benchmarking Hub | |
| Arena Hard Prompts | – | 1489#50 | – | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 77.6%#220 | 91.1%#50 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 27.9%#169 | 34.5%#126 | – | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | – | 88.6%#33 | – | Vals AI | |
| Kagi LLM Benchmark | – | 75.0%#12 | 44.8%#95 | Kagi LLM Benchmark | |
| ARC-AGI-1 | – | 89.5%#83 | – | ARC Prize | |
| ARC-AGI-2 | – | 65.1%#74 | – | ARC Prize | |
| ARC-AGI-3 | – | 0.1%#64 | – | ARC Prize | |
| CritPt | 0.3%#271 | 6.6%#145 | – | Artificial Analysis | |
| Chess Puzzlesnot in index | – | 29.0%#69 | – | Epoch AI Benchmarking Hub | |
| CL-benchnot in indexeffort not stated | – | – | 22.2%#3 | Epoch AI Benchmarking Hub | |
| CL-bench Lifenot in indexeffort not stated | – | – | 11.9%#12 | Epoch AI Benchmarking Hub | |
| LMCAnot in indexeffort not stated | – | – | 38.7%#112 | Epoch AI Benchmarking Hub | |
| DTBenchnot in indexeffort not stated | – | – | 90.1%#82 | Epoch AI Benchmarking Hub | |
| ForecastBenchnot in index | – | 64.4%#25 | 63.3%#44 | ForecastBench |
Coding
| Benchmark | thinking | not stated | Source | Trend |
|---|---|---|---|---|
| WeirdMLeffort not stated | – | 52.3%#66 | WeirdML | |
| Arena Coding | 1511#58 | – | Arena (LMArena) | |
| Arena WebDev | 1376#101 | – | Arena (LMArena) | |
| LiveCodeBench | 84.3%#38 | – | Vals AI | |
| SWE-bench (Vals)not in index | 72.2%#54 | – | Vals AI | |
| Code Migrationnot in index | 0.3%#74 | – | Vals AI | |
| Vibe Code Bench v1.1not in index | 4.1%#102 | – | Vals AI | |
| ALE-Benchnot in indexeffort not stated | – | 1150.3#49 | Epoch AI Benchmarking Hub |
Agents & tools
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| Terminal-Bencheffort not stated | – | – | 57.3%#24 | Terminal-Bench | |
| Terminal-Bench Hard | 16.7%#184 | 37.9%#62 | – | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 59.9%#180 | 93.0%#46 | – | Artificial Analysis | |
| Terminal-Bench 2.1 (Vals) | – | 44.2%#62 | – | Vals AI | |
| HiL-Bencheffort not stated | – | – | 20.0%#26 | Scale AI SEAL | |
| APEX-Agents (AA)not in index | – | 14.2%#26 | – | Artificial Analysis | |
| Finance Agent v2not in index | – | 28.5%#71 | – | Vals AI | |
| Harvey's Legal Agent Benchmark v1not in index | – | 0.0%#61 | – | Vals AI | |
| Legal Research Benchnot in index | – | 13.9%#68 | – | Vals AI | |
| Tax Agent Benchnot in index | – | 7.0%#63 | – | Vals AI | |
| Terminal-Bench 2.0 (Vals)not in index | – | 40.5%#32 | – | Vals AI | |
| Vending-Bench 2not in indexeffort not stated | – | – | 4662.9#36 | Andon Labs |
Maths
| Benchmark | thinking | Source | Trend |
|---|---|---|---|
| FrontierMath Tiers 1–3 | 44.9%#57 | Epoch AI Benchmarking Hub | |
| FrontierMath Tier 4 | 17.1%#56 | Epoch AI Benchmarking Hub | |
| OTIS Mock AIME | 92.2%#66 | Epoch AI Benchmarking Hub | |
| ProofBench | 14.0%#61 | Vals AI | |
| AIME (Vals) | 96.5%#6 | Vals AI | |
| Arena Maths | 1465#59 | Arena (LMArena) |
Knowledge
| Benchmark | no reasoning | thinking | Source | Trend |
|---|---|---|---|---|
| SimpleQA Verified | – | 30.2%#67 | Epoch AI Benchmarking Hub | |
| AA-Omniscience | -44.6#349 | 14.8#83 | Artificial Analysis | |
| MMLU-Pro | – | 86.3%#42 | Vals AI | |
| LegalBench | – | 77.7%#103 | Vals AI | |
| CorpFin | – | 63.7%#42 | Vals AI | |
| TaxEval | – | 74.1%#40 | Vals AI | |
| MedQA | – | 94.5%#16 | Vals AI | |
| CaseLaw v2not in index | – | 54.5%#38 | Vals AI | |
| Excel Modeling Benchmarknot in index | – | 11.7%#70 | Vals AI | |
| MedCodenot in index | – | 32.2%#88 | Vals AI | |
| MedScribenot in index | – | 63.4%#99 | Vals AI | |
| AA-Omniscience: accuracynot in index | 26.7%#213 | 27.5%#205 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 2.7%#545 | 82.6%#10 | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | thinking | Source | Trend |
|---|---|---|---|---|
| IFBench | 49.3%#178 | 81.2%#5 | Artificial Analysis | |
| Arena Instruction Followingnot in index | – | 1447#67 | Arena (LMArena) |
Human preference
| Benchmark | thinking | Source | Trend |
|---|---|---|---|
| Arena Text | 1472#46 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1446#50 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1479#42 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1466#67 | Arena (LMArena) |
Multimodal
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| Arena Vision | – | 1262#55 | – | Arena (LMArena) | |
| MMMU-Pro | 64.9%#179 | 74.6%#110 | – | Artificial Analysis | |
| MMMU-Pro (Vals)not in index | – | 83.5%#32 | – | Vals AI | |
| MortgageTaxnot in index | – | 45.4%#82 | – | Vals AI | |
| SAGEnot in index | – | 38.2%#62 | – | Vals AI | |
| Blueprint-Bench 2not in indexeffort not stated | – | – | 0.0%#29 | Epoch AI Benchmarking Hub |
Long context
| Benchmark | no reasoning | thinking | Source | Trend |
|---|---|---|---|---|
| AA-LCR | 23.3%#419 | 69.0%#232 | Artificial Analysis | |
| Arena Documentnot in index | – | 1439#35 | Arena (LMArena) |
Composite
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in indexeffort not stated | – | – | 152.1#49 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index v4.3.2 | 14.2#297 | 25.7#161 | – | Artificial Analysis | |
| Vals Multimodal Indexnot in index | – | 39.1%#30 | – | Vals AI |
Where it wins
Benchmarks where this configuration ranks in the top five of every configuration measured.
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| SpaceXAI | 82 tok/s | 0.69 s | $1.25 | $2.50 | 2M | – |
| SpaceXAI | 77 tok/s | 0.71 s | $1.25 | $2.50 | 2M | – |
Price history
The price above, per 1M tokens, every day it changed.
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.200 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0013 | $0.0009 | 19.8 s |
| Summarize a 30-page report | 12,000 / 600 | $0.017 | $0.0071 | 22.5 s |
| Code edit | 6,000 / 1,500 | $0.011 | $0.0065 | 30.6 s |
| Agentic coding session | 60,000 / 4,000 | $0.085 | $0.038 | 52.9 s |
| Structured extraction | 2,000 / 200 | $0.0030 | $0.0014 | 18.9 s |
More from SpaceXAI
See also
Questions people ask
- What are Grok 4.20's benchmark scores?
- As of 24 Oct 2026, Grok 4.20 (thinking effort) scores 58.6 ±3.8 on the BenchLeader Index, strongest in instruction following (79) and weakest in agents & tools (50). That rests on 31 independent benchmark results, and every score on this page links to its publisher.
- How does Grok 4.20 rank?
- Grok 4.20 (thinking effort) is #148 of 785 ranked configurations on 24 Oct 2026. 231 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- How much does Grok 4.20 cost per 1M tokens?
- As of 24 Oct 2026, Grok 4.20 costs $1.25 per 1M input tokens and $2.50 per 1M output tokens, $1.56 blended (three input tokens per output token). The list price last changed on 18 Oct 2026.
- When was Grok 4.20 released?
- SpaceXAI released Grok 4.20 on 17 Feb 2026. BenchLeader has results for it at 3 reasoning-effort settings.
Data as of 24 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “Grok 4.20: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/grok-4-20.