tokeven4 models across 3 tiers
Grok API pricing in 2026
xAI has the Grok family across three tiers: flagship (Grok 4.6, Grok 4.5), mid-tier (Grok 4.3) and fast (Grok Build 0.1, the family's coding model and the cheapest Grok row on output). Two things make Grok price differently from the other three providers. First, cached input is quoted as an absolute rate per model rather than a family-wide discount, so the effective cache-read discount ranges from 75% on Grok 4.6 to 85% on Grok 4.5. Grok 4.5 is cheaper than Grok 4.6 on cache-heavy work despite identical list prices. Second, every Grok model re-rates the entire request at 2x once the prompt reaches 200,000 tokens, so long-context work costs double the headline rate. There is no cache-write surcharge.
Models
4
Cheapest input
$1.00/MTok
Cache discount
80% off
Max context
1,000K
All Grok models
| Model | Tier | Input $/MTok | Output $/MTok | Context | Best task | Released |
|---|---|---|---|---|---|---|
| Grok Build 0.1 | Fast / Budget | $1.00 | $2.00 | 256K | Not benchmarked | May 2026 |
| Grok 4.3 | Mid-tier | $1.25 | $2.50 | 1,000K | Not benchmarked | Mar 2026 |
| Grok 4.6 | Flagship | $2.00 | $6.00 | 500K | Not benchmarked | Aug 2026 |
| Grok 4.5 | Flagship | $2.00 | $6.00 | 500K | Not benchmarked | Jul 2026 |
Grok Build 0.1: xAI's coding-specialised model and the cheapest Grok row on output. Released in early access; xAI publishes release notes at month granularity only.
Grok 4.3: xAI publishes no release date for this model; dated to the March 4.20-generation launch whose pricing and 1M context window it shares. Priced from the live models table, which is what matters for costing traffic.
Grok 4.5: Same list price as Grok 4.6 but a cheaper cached rate ($0.30/MTok vs $0.50), so cache-heavy workloads are cheaper here than on 4.6. xAI publishes release notes at month granularity only; dated to the start of the month.
Grok cost by use case
Estimated cost per API call across common workloads, using default token counts for each use case.
Grok quality benchmarks by task
Confidence scores (0-100) across all task types. Higher is better.
Confidence scores are a third-party benchmark snapshot from week 29 of 2026. This snapshot is 8 weeks old, so treat the scores as directional and validate against your own evals.
| Task | 0.1 | 4.3 | 4.6 | 4.5 |
|---|---|---|---|---|
| Code Generation | - | - | - | - |
| Code Review | - | - | - | - |
| Summarization | - | - | - | - |
| Q&A | - | - | - | - |
| Extraction | - | - | - | - |
| Reasoning | - | - | - | - |
| Classification | - | - | - | - |
| Creative Writing | - | - | - | - |
Recent Grok pricing changes
Grok 4.6 input launched
$2.00/MTok at launch
Grok 4.5 input launched
$2.00/MTok at launch
Grok Build 0.1 input launched
$1.00/MTok at launch
Grok 4.3 input launched
$1.25/MTok at launch
Compare Grok with other providers
Head-to-head comparisons
Track your Grok spend
See exactly how much you spend on Grok models and where you can save. Two-line wrapper install, no API keys shared.