benchrRouting note: Grok 4.3 is now xAI's default — legacy model slugs redirect to it and bill at its rates. Details in the default-and-pricing story.
Pricing breakdown
| Tier | Rate / 1M tokens |
|---|---|
| Standard input | $1.25 |
| Standard output | $2.50 |
| Context window | 256,000 tokens |
The unusual output pricing
Most frontier models price output at 4–8× the input rate: GPT-5 charges $10/1M output vs $1.25/1M input (8:1). Claude Opus 4.8 charges $25/1M output vs $5/1M input (5:1). Grok 4.3's $2.50/1M output vs $1.25/1M input is a 2:1 ratio — a structural cost advantage for output-heavy tasks. If your workflow generates 5M output tokens per month, the cost difference is: Grok 4.3 at $12.50 versus GPT-5 at $50. For content generation, summarization, documentation, and report writing, this pricing structure is meaningfully different.
Native real-time web search
Grok 4.3 integrates web search directly into its reasoning via the xAI API — not as a tool call that returns documents, but as an architecture feature that lets the model fetch current information as part of generation. This enables: factual accuracy on recent events without RAG infrastructure; current market data, news, and research without knowledge cutoff gaps; real-time information in production apps without building a separate retrieval pipeline. For use cases where information freshness matters, Grok 4.3 removes engineering overhead that other models require.
Cost scenarios
At 10M input + 10M output per month — an output-balanced workload: Grok 4.3 costs $12.50 + $25 = $37.50/month. GPT-5 at the same volume: $12.50 + $100 = $112.50/month — GPT-5 is 3× more expensive on this output-heavy profile. Claude Sonnet 4.6 at the same volume: $30 + $150 = $180/month — Grok 4.3 is 4.8× cheaper. For pure output volume, the economics are compelling.
Use-case fit
Best for: Long-form content generation; report and documentation pipelines where output volume is high; research tasks that benefit from real-time web access; applications where the OpenAI-compatible API format allows easy integration; output-heavy workloads where GPT-5's $10/1M output rate is a cost concern.
Skip if: SWE-bench coding performance is the primary benchmark — Grok 4.3's coding benchmarks trail Claude Opus 4.8 and GPT-5.5. Also skip for teams deeply integrated with Anthropic or Google tooling where the xAI ecosystem is unfamiliar.
Decision checklist
Calculate your input-to-output token ratio in production. If your workload generates 2+ output tokens per input token, Grok 4.3's cheap output pricing creates meaningful savings. If your workload is input-heavy (long documents, short answers), the output pricing advantage diminishes.
For real-time search: evaluate whether your current RAG pipeline complexity justifies migration. If you're building retrieval infrastructure from scratch, native search support may reduce time-to-production significantly.