BlogLLM Benchmark Scores as Published, Side by Side
Grok 4.5: Benchmark Scores as Published
Every result the catalog holds for Grok 4.5: 41 tests in 10 areas, from 6 pages. 38 printed by xAI itself, the rest by the company named beside each number. Each number is as printed; nothing is ranked or rescored.
Developer xAIReleased 2026-07-16Context window 500KWeights ClosedData version 2026103001, as of 2026-10-30
Knowledge Work
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| GDPval-AA v2 | 1526 | xAI, Introducing Grok 4.6checked 2026-10-15; printed on 1 more page | Effort: high |
| APEX-Agents | 47.1% | xAI, Introducing Grok 4.6checked 2026-10-15 | Effort: high |
| Harvey's Legal Agent Benchmark | 12.9% | xAI, Introducing Grok 4.6checked 2026-10-15 | Effort: high |
| AA-Briefcase | 1313 | xAI, Introducing Grok 4.6checked 2026-10-15 | Effort: high |
Agentic Coding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| SWE-Bench Pro | 64.7% | xAI, Grok 4.5checked 2026-10-15; printed on 1 more page | - |
| Terminal-Bench 2.1 | 83.3% | xAI, Grok 4.5checked 2026-10-15; printed on 1 more page | - |
| DeepSWE v1.1 | 53% | xAI, Grok 4.5checked 2026-10-15; printed on 2 more pages | Harness: mini-swe-agent (Datacurve run) |
| Terminal-Bench 4.0 | 12.4% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| SWE-Marathon | 29% | xAI, Grok 4.5checked 2026-10-15 | - |
| Terminal-Bench 3.0 | 15.7% | xAI, Introducing Grok 4.6checked 2026-10-15 | Effort: high |
| FrontierCode v1.1 Extended | 56.6% | xAI, Introducing Grok 4.6checked 2026-10-15 | Effort: high |
| CursorBench 3.2.0 | 66.7% | xAI, Introducing Grok 4.6checked 2026-10-15 | Effort: high |
| DeepSWE v1.0 | 62% | xAI, Grok 4.5checked 2026-10-15 | Harness: each provider's own harness (Artificial Analysis run) |
| APEX-SWE | 53.6% | xAI, Introducing Grok 4.6checked 2026-10-15 | Effort: high |
ML Engineering
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| MLE-bench | 43.2% | Google, Gemini 3.6 Flash model cardchecked 2026-10-15 | - |
Science and Math
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| ProtocolQA (open-ended) | 87% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| BixBench | 93.8% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| LAB-Bench (practical MCQ) | 71.1% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
Long Context
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| MRCR v2 8-needle 128K | 81.4% | Google, Gemini 3.6 Flash model cardchecked 2026-10-15 | Context: 128k (average) |
Multimodal Understanding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CharXiv Reasoning (no tools) | 81.6% | Google, Gemini 3.6 Flash model cardchecked 2026-10-15 | Tools: none |
Cybersecurity
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CyberGym | 79% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| HackerBench v0.3 benign refusallower is better | 0% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| HackerBench v0.3 harmful and dual-use compliancelower is better | 7.8% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| CVE-Bench reward | 35.2% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| WMDP-Cyber | 83.2% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
Biosecurity
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Biosecurity VCT (xAI internal) | 44.1% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| BioUseBench dual-use refusal | 83.3% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Virology Capabilities Test (VCT) | 65.5% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| WMDP-Bio | 90.9% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| WMDP-Chem | 87.3% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
Safety and Alignment
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CBRN bio refusal recall | 97.9% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| CBRN chem refusal recall | 96.7% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Child safety compliancelower is better | 0% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| FORTRESS-RN refusal accuracy | 97.9% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| General refusals compliancelower is better | 1.1% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| MASK-Rectified dishonestylower is better | 0.67% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Self-harm refusals compliancelower is better | 0.5% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Standard jailbreaks compliancelower is better | 0.73% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| StrongReject compliancelower is better | 1.5% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Sycophancy ratelower is better | 0.01% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
Preference and Composite Rankings
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Artificial Analysis Intelligence Index | 56 | xAI, Introducing Grok 4.6checked 2026-10-15 | Effort: high |
Price
API list prices in US dollars per one million tokens, as printed on the developer's pricing page on the day checked. Cost per task is measured by Vals AI.
| Price | Value | Printed by | Setup as printed |
|---|---|---|---|
| Input | $2 | xAI, xAI API pricingchecked 2026-10-15 | Context: up to 200,000 tokens; Long context: $4 per 1M tokens above 200,000 tokens |
| Cached input | $0.3 | xAI, xAI API pricingchecked 2026-10-15 | Context: up to 200,000 tokens; Long context: $0.6 per 1M tokens above 200,000 tokens |
| Output | $6 | xAI, xAI API pricingchecked 2026-10-15 | Context: up to 200,000 tokens; Long context: $12 per 1M tokens above 200,000 tokens |
| Cost per taskVals AI, USD, lower is better | $4.1 | Vals AI, Vibe Code Bench v1.1 leaderboardchecked 2026-10-27 | Harness: OpenHands |
More from xAI
Every number here is read from the same catalog file as the benchmark table, data version 2026103001. Where one page printed the model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.