Grok 4.3
xAI's rapidly climbing frontier model with real-time X/data hooks and competitive Arena scores.
When to use it
Current-events chat, witty assistants, and X-integrated products.
Watch-outs
Independent coding/science benches are less consistently published than peers.
Benchmarks
Human preference ranking from blind pairwise chats. Higher is better; top frontier models cluster within ~50–80 Elo.
Artificial Analysis composite across agents, coding, science, and general evaluations (v4.1 weighting).
Percent of real GitHub issues resolved end-to-end. Strong signal for agentic coding usefulness.
PhD-level science questions. Separates frontier reasoning models better than saturated knowledge tests.
Harder multi-choice knowledge/reasoning suite than classic MMLU.
Frontier closed-ended academic difficulty across many domains.