Grok 4.5 vs Grok 4.6
Different efforts: high · medium · Compare all at high
AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation
Grok 4.557.9
Grok 4.663.2
GDPval-AA (work tasks) 2.1 GDPval-AA (work tasks) 2.1: Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work
Grok 4.544.2
Grok 4.655.6
τ³-Banking · AA τ³-Banking · AA: Agents retrieve banking knowledge and execute tools to resolve customer-support workflows
Grok 4.542.1
Grok 4.644.3
AA Analyst Agent AA Analyst Agent: Data-analyst tasks: the model explores data with tools and reports findings
Grok 4.535
Grok 4.6 · high41.3
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 4.510.6
Grok 4.613.1
AA-Briefcase 1.1 · Elo AA-Briefcase 1.1 · Elo: Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating
Grok 4.51,280.6
Grok 4.61,483.2
AA-LCR (long context) 1.1 AA-LCR (long context) 1.1: Questions that need reading and connecting facts across very long documents (around 100K tokens)
Grok 4.579.3
Grok 4.681
DeepSWE 1.1 DeepSWE 1.1: Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition
Grok 4.553.8
Grok 4.667.5
EnterpriseOps-Gym-AA EnterpriseOps-Gym-AA: Agents complete stateful enterprise workflows in eight domains; final database state determines success
Grok 4.540.8
Grok 4.6 · high48.3
Furniture Assembly Furniture Assembly: Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step
Grok 4.522.5
Grok 4.6 · xhigh40
τ³-bench · banking · 1.0.1 τ-bench official · 3-banking-1.0.1: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools
Grok 4.547.9
Grok 4.6—
Terminal-Bench Science 0.1 Terminal-Bench Science 0.1: Agents complete expert-authored research workflows across five scientific domains in a terminal
Grok 4.5—
Grok 4.6 · high6.2
EBR Bench EBR Bench: Agents learn to play Earthborne Rangers across repeated playthroughs
Grok 4.5—
Grok 4.6 · xhigh30.5
| Benchmark | Grok 4.5 | Grok 4.6 |
|---|---|---|
| AutomationBench-AA | 57.9 | 63.2 |
| GDPval-AA (work tasks) 2.1 | 44.2 | 55.6 |
| τ³-Banking · AA | 42.1 | 44.3 |
| AA Analyst Agent | 35 | 41.3 · high |
| Terminal-Bench 4.0 | 10.6 | 13.1 |
| AA-Briefcase 1.1 · Elo | 1,280.6 | 1,483.2 |
| AA-LCR (long context) 1.1 | 79.3 | 81 |
| DeepSWE 1.1 | 53.8 | 67.5 |
| EnterpriseOps-Gym-AA | 40.8 | 48.3 · high |
| Furniture Assembly | 22.5 | 40 · xhigh |
| τ³-bench · banking · 1.0.1 | 47.9 | — |
| Terminal-Bench Science 0.1 | — | 6.2 · high |
| EBR Bench | — | 30.5 · xhigh |
Specs
| Spec | Grok 4.5 | Grok 4.6 |
|---|---|---|
| Input, per 1M | $2 | $2 |
| Output, per 1M | $6 | $6 |
| Cost per task? | $0.88 | $1.14 |
| Speed, tok/s | 63 | 68 |
| Context | 500K | 500K |
| Weights | Closed | Closed |
| Released | Jul 2026 | Aug 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 4.5 → Grok 4.6
Nothing known.
Grok 4.6 → Grok 4.5
Nothing known.