Compare Models

Grok 4.5

Grok 4.5 is a model from SpaceXAI, released July 16, 2026. It ranks #37 of 45 models on the Vals Index with 44.72%. Its best result is #2 of 35 on SkillsBench.

Release Date: Jul 16, 2026

Developer SpaceXAI  🇺🇸
Context Window 500k
Max Output Tokens 500k
Token Costs (in/out) $2.00/6.00
Weights Private
Input Modalities

Accuracy

44.72 % ± 1.17

Cost / Test (Vals Index)

$ 2.459

Latency

15 min 44 s

Vals Index
BenchmarksAccuracyRankings
Vals Index

44.72%

±1.17
37/45
Code Migration

36.60%

±4.14
31/74
CyberBench v1.1

71.01%

±5.79
17/44
EMB

52.91%

±3.11
48/72
Finance Agent (v2)

48.35%

±0.46
46/76
Time Horizon Index: KSP

6.33%

±0.00
11/13
Legal Research Bench

37.98%

±3.37
35/75
MedCode

43.29%

±2.31
44/105
MedScribe

86.88%

±1.94
17/107
MortgageTax

61.80%

±0.97
59/98
ProofBench v1.1

31.00%

±4.65
34/48
SAGE

34.97%

±3.32
70/90
SRE Bench

0.76%

±0.54
27/32
Tax Agent Bench

64.32%

±3.16
21/67
TaxEval v2

71.67%

±0.89
76/145
Vibe Code Bench v1.1

69.00%

±4.48
39/110
Harvey's Legal Agent Benchmark

12.92%

±2.53
8/76
GPQA Diamond

92.93%

±1.29
13/138
IOI

40.56%

±3.95
32/42
LiveCodeBench

87.35%

±0.96
15/143
LegalBench

85.97%

±0.41
17/149
MMLU Pro

89.22%

±0.31
14/138
MMMU Pro

61.79%

±1.17
83/93
ProgramBench

0.00%

±0.00
40/62
SkillsBench

66.03%

±4.60
2/35
SWE-bench

86.60%

±1.52
13/88
Terminal-Bench 4.0

8.59%

±1.34
35/45
Proprietary BenchmarksAcademic BenchmarksIndustry Partners
Vals
Default Provider : SpaceXAI
Temperature: 0.7
Top P: 0.95
Top K: Default
Max Output Tokens: 500,000
Reasoning Effort: high

Compared with

Grok 4.5 vs Qwen 3.8 Max 25 shared benchmarks Grok 4.5 vs DeepSeek V4 Pro 0813 23 shared benchmarks Grok 4.5 vs Claude Sonnet 4.6 18 shared benchmarks Grok 4.5 vs Gemini 3.5 Flash 23 shared benchmarks Grok 4.5 vs GLM 5.2 17 shared benchmarks Grok 4.5 vs GLM 5.3 Flash 21 shared benchmarks Grok 4.5 vs DeepSeek V4 Flash 0731 22 shared benchmarks Grok 4.5 vs Gemini 3.6 Flash 21 shared benchmarks

Updates

Jul 8, 2026

We evaluated SpaceXAI’s new Grok 4.5 across our benchmark suite.

The model has a 500k-token context window and supports up to 500k output tokens. Evaluations were run with reasoning effort set to “high”, temperature 0.7, and top-p 0.95.

Congrats to the SpaceXAI team on the release!