xAIGrok 4.20
Best configuration ranks #135 of 610 on the BenchLeader Index at 57.9 ±7.5. Last measured 16 Apr 2026. Released 17 Feb 2026.
- Blended price
- $1.56/M
- $1.25 in · $2.50 out
- Output speed
- 104 tok/s
- First answer
- 19 s
- first token 0.57 s
- Context
- 1M
- Overall index58
- Reasoning42
- Coding52
- Agents & tools57
- Knowledge71
- Instruction following79
- Multimodal59
- Long context61
- Composite61
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. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | 47.9 | #343 | 92 tok/s | 0.59 s | $0.0013 | Agents & tools 53 · Composite 47 · Instruction following 50 · Knowledge 44 · Long context 38 · Multimodal 49 |
| thinking | 54.3 | #201 | 104 tok/s | 19 s | $0.0013 | Agents & tools 34 · Coding 55 · Human preference 67 · Knowledge 53 · Maths 52 · Multimodal 61 · Reasoning 62 |
| defaultbest | 57.9 | #135 | 104 tok/s | 19 s | $0.0013 | Agents & tools 57 · Coding 52 · Composite 61 · Instruction following 79 · Knowledge 71 · Long context 61 · Multimodal 59 · Reasoning 42 |
Benchmark results
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 | default | Source |
|---|---|---|---|---|
| GPQA Diamond | – | 89.3%#44 | – | Epoch AI Benchmarking Hub |
| LMArena Hard Prompts | – | 1488#43 | – | LMArena |
| GPQA Diamond (AA)not in index | 78.5%#203 | – | 91.1%#50 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 27.9%#121 | – | 34.5%#82 | Artificial Analysis |
| GPQA Diamond (Vals)not in index | – | 88.6%#33 | – | Vals AI |
| Kagi LLM Benchmark | – | 75.0%#12 | 44.8%#94 | Kagi LLM Benchmark |
| ARC-AGI-1 | – | 89.5%#55 | – | ARC Prize |
| ARC-AGI-2 | – | 65.1%#47 | – | ARC Prize |
| ARC-AGI-3 | – | 0.1%#35 | – | ARC Prize |
Coding
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| WeirdML | – | – | 52.3%#63 | WeirdML |
| LMArena Coding | – | 1510#45 | – | LMArena |
| LMArena WebDev | – | 1373#81 | – | LMArena |
| LiveCodeBench | – | 84.3%#38 | – | Vals AI |
| IOI | – | 30.2%#16 | – | Vals AI |
| SWE-bench (Vals)not in index | – | 72.2%#54 | – | Vals AI |
Agents & tools
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| Terminal-Bench | – | – | 57.3%#20 | Terminal-Bench |
| Terminal-Bench Hard | 22.0%#149 | – | 40.9%#51 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 69.6%#152 | – | 96.5%#14 | Artificial Analysis |
| Terminal-Bench 2.1 (Vals) | – | 44.2%#52 | – | Vals AI |
| HiL-Bench | – | – | 20.0%#13 | Scale AI SEAL |
Maths
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| FrontierMath Tiers 1–3 | – | 44.9%#42 | – | Epoch AI Benchmarking Hub |
| FrontierMath Tier 4 | – | 17.1%#42 | – | Epoch AI Benchmarking Hub |
| OTIS Mock AIME | – | 92.2%#51 | – | Epoch AI Benchmarking Hub |
| ProofBench | – | 14.0%#46 | – | Vals AI |
| AIME (Vals) | – | 96.5%#6 | – | Vals AI |
Knowledge
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| SimpleQA Verified | – | 30.2%#57 | – | Epoch AI Benchmarking Hub |
| AA-Omniscience | -42.6#272 | – | 14.8#53 | Artificial Analysis |
| MMLU-Pro | – | 86.3%#42 | – | Vals AI |
| LegalBench | – | 77.7%#98 | – | Vals AI |
| CorpFin | – | 63.7%#42 | – | Vals AI |
| TaxEval | – | 74.1%#40 | – | Vals AI |
| MedQA | – | 94.5%#16 | – | Vals AI |
Instruction following
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| IFBench | 49.3%#175 | – | 82.9%#2 | Artificial Analysis |
Human preference
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| LMArena Text | – | 1472#36 | – | LMArena |
Multimodal
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| LMArena Vision | – | 1263#39 | – | LMArena |
| MMMU-Pro | 64.9%#146 | – | 74.6%#82 | Artificial Analysis |
Long context
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| AA-LCR | 25.3%#339 | – | 69.0%#176 | Artificial Analysis |
Composite
| Benchmark | no reasoning | thinking | default | Source |
|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | 152.0#36 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | 14.6#228 | – | 25.7#113 | Artificial Analysis |
| Vals Indexnot in index | – | 17.6#50 | – | 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 | 65 tok/s | 0.57 s | $1.25 | $2.50 | 2M | – |
| SpaceXAI Priority (ZDR) | 53 tok/s | 0.50 s | $2.50 | $5.00 | 2M | – |
| SpaceXAI (ZDR) | 43 tok/s | 0.89 s | $1.25 | $2.50 | 2M | – |
Price history
Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.
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. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0013 | – | 22.3 s |
| Summarise a 30-page report | 12,000 / 600 | $0.017 | – | 25.2 s |
| Code edit | 6,000 / 1,500 | $0.011 | – | 33.8 s |
| Agentic coding session | 60,000 / 4,000 | $0.085 | – | 58.0 s |
| Structured extraction | 2,000 / 200 | $0.0030 | – | 21.3 s |
See also
Data as of 9 Sept 2026. Compare these configurations.