vs
0:4
overall capability wins (0 ties across 4 scored). Grok 4.7 leads overall.
- Reasoning0:2
- Coding0:1
- Pricing0:3
- Specs0:1
Verdict
Grok 4.7 leads coding (widest gap: +20.2 on SciCode); Grok 4.7 offers ~3.8x larger context (500K vs 131K); Grok 4.7 is ~2.0x cheaper on a blended token basis than Grok 3.
- Grok 4.7 leads coding (widest gap: +20.2 on SciCode)
- Grok 4.7 offers ~3.8x larger context (500K vs 131K)
- Grok 4.7 is ~2.0x cheaper on a blended token basis than Grok 3
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Grok 3
- $6.75
- Grok 4.7
- $3.50
Grok 4.7 is about 1.9x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
xAI
xAI
Family
Grok
Grok
License
Proprietary
Proprietary
Open weights
No
No
Release
Feb 17, 2025
Sep 21, 2026
Knowledge cutoff
-
-
API / provider
xAI
xAI
Modalities
text, image → text
text, image → text
Specs
Context window
131K
500K
Max output
33K
-
Parameters
—
—
Pricing
Input $/1M
$3
$2
Output $/1M
$15
$6
Blended $/1M (3∶1)
$6
$3
Speed
tok/s
75
-
TTFT (s)
0.45
-
Reasoning
MMLU-Pro
79.9
—
GPQA Diamond
69.3
—
Humanity's Last Exam
4.1
43.0
AIME 2025
58.0
—
MATH-500
87.0
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
28.9
—
AA-LCR v1.1
58.0
77.0
CritPt
—
18.0
MMMU-Pro
—
—
IFBench
46.9
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
42.5
—
Terminal-Bench 2.1
—
—
Aider Polyglot
53.3
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
36.8
57.0
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
—
Terminal-Bench 4.0
—
26.0
CursorBench 4.0
—
46.3
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,412
—
| Metric | Grok 3 | Grok 4.7 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | xAI | xAI | — |
| Family | Grok | Grok | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | Feb 17, 2025 | Sep 21, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | xAI | xAI | — |
| Modalities | text, image → text | text, image → text | — |
| Specs | |||
| Context window | 131K | 500K | B +369K |
| Max output | 33K | - | — |
| Parameters | — | — | — |
| Pricing | |||
| Input $/1M | $3 | $2 | B +$1 |
| Output $/1M | $15 | $6 | B +$9 |
| Blended $/1M (3∶1) | $6 | $3 | B +$3 |
| Speed | |||
| tok/s | 75 | - | — |
| TTFT (s) | 0.45 | - | — |
| Reasoning | |||
| MMLU-Pro | 79.9 | — | — |
| GPQA Diamond | 69.3 | — | — |
| Humanity's Last Exam | 4.1 | 43.0 | B +38.9 pts |
| AIME 2025 | 58.0 | — | — |
| MATH-500 | 87.0 | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 28.9 | — | — |
| AA-LCR v1.1 | 58.0 | 77.0 | B +19.0 pts |
| CritPt | — | 18.0 | — |
| MMMU-Pro | — | — | — |
| IFBench | 46.9 | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | 42.5 | — | — |
| Terminal-Bench 2.1 | — | — | — |
| Aider Polyglot | 53.3 | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 36.8 | 57.0 | B +20.2 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | 26.0 | — |
| CursorBench 4.0 | — | 46.3 | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,412 | — | — |
Grok 3
- Context
- 131K / 33K out
- Parameters
- —
- Price
- $3 / $15
- Speed
- 75 tok/s · 0.45s TTFT
- Modalities
- text, image → text
- License
- Proprietary
Previous-gen Grok flagship; retained for API compatibility and real-time X context.
Grok 4.7
- Context
- 500K
- Parameters
- —
- Price
- $2 / $6
- Speed
- —
- Modalities
- text, image → text
- License
- Proprietary
SpaceXAI's September 2026 frontier model for coding, agentic tasks, and knowledge work. Its API base rates are $2/$6 per 1M input/output tokens, with higher rates above 200K context; the context window is 500K. Recorded Artificial Analysis evaluations use xhigh reasoning effort.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.