Grok 4.6 vs Grok 4.7
Different efforts: medium · low · Compare all at high
Humanity's Last Exam Humanity's Last Exam: Very hard expert questions across many fields, written to stump today’s models
Grok 4.642.1
Grok 4.739.2
Terminal-Bench 4.0 Terminal-Bench 4.0 · source setting 2: Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite
Grok 4.613.1
Grok 4.716.2
AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation
Grok 4.663.2
Grok 4.756.7
Arena Text · creative writing Arena Text · creative writing: Human pairwise preferences in the stated text vote pool
Grok 4.6 · high1,442
Grok 4.7 · xhigh1,424
MMMU-Pro (vision) MMMU-Pro (vision): College-level questions that require understanding images: charts, diagrams, photos
Grok 4.6—
Grok 4.778.2
| Benchmark | Grok 4.6 | Grok 4.7 |
|---|---|---|
| Humanity's Last Exam | 42.1 | 39.2 |
| Terminal-Bench 4.0 | 13.1 | 16.2 |
| AutomationBench-AA | 63.2 | 56.7 |
| Arena Text · creative writing | 1,442 · high | 1,424 · xhigh |
| MMMU-Pro (vision) | — | 78.2 |
Specs
| Spec | Grok 4.6 | Grok 4.7 |
|---|---|---|
| Input, per 1M | $2 | $2 |
| Output, per 1M | $6 | $6 |
| Cost per task? | $1.14 | $1.25 |
| Speed, tok/s | 68 | 80 |
| Context | 500K | 500K |
| Weights | Closed | Closed |
| Released | Aug 2026 | Sep 2026 |
| See images | ✓ | ✓ |
| Read PDFs | Partial | Partial |
| Hear audio | — | — |
| Watch video | — | — |
| Call tools | ✓ | ✓ |
| Force a tool call | Partial | Partial |
| Return strict JSON | ✓ | ✓ |
| Search the web | ? | ✓ |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
Grok 4.6 → Grok 4.7
Reasoning cannot be disabled›
Reasoning cannot be disabled; choose low, medium, high or xhigh effort.
Penalties and stop return error›
presencePenalty, frequencyPenalty and stop cannot be used with reasoning models; requests containing them return an error.
Set prompt_cache_key›
Set prompt_cache_key or the Chat Completions x-grok-conv-id header to keep a conversation on the same server and improve cache hits.
Pass reasoning items back unchanged · Responses›
Responses always returns encrypted reasoning even without include; pass the reasoning items back unchanged in the next request.
10% premium · US regional endpoint›
The US regional endpoint keeps inference in the United States and adds a 10% token-price premium.
Grok 4.7 → Grok 4.6
Nothing known.