$ cat wiki/models/grok-v9-medium.md
Grok V9-Medium
modelupdated 2026-07-21created 2026-06-23
xAI's coding-focused foundation model — 1.5-trillion parameters, ~3× larger than the prior production Grok. Training completed late May 2026 (announced by Elon Musk 2026-05-25); shipped to consumers 2026-06-16. Not Grok 5.
Spec
| Attribute | Value |
|---|---|
| Developer | xai |
| Released | 2026-06-16 |
| Announced | 2026-05-25 (Musk announced training complete) |
| Context window | unknown |
| Pricing | unknown |
| License | unknown |
| Availability | Consumer Grok product on X and SuperGrok; grok-v9-medium not in xAI API docs as of 2026-06-19 (consumer-first launch) |
| Parameters | ~1.5T (≈3× the prior production model) |
| Training focus | trained on Cursor developer workflows — targeting Claude's coding-benchmark lead |
Release Date
2026-06-16 (consumer). API availability lagging.
Benchmarks
No independent benchmark numbers published at launch; xAI positions it against Claude's coding lead. (To verify on next ingest.)
Use Cases
- Agentic / assisted coding (Cursor-style workflows) — xAI's bid for the coding-model crown.
Compared To
| Model | Org | Focus | Notes |
|---|---|---|---|
| Grok V9-Medium | xAI | coding | 1.5T; Cursor-trained; consumer-first |
| Claude Fable 5 | Anthropic | frontier | SWE-Bench Pro 80.3% (suspended 2026-06-12) |
| Grok 5 (training) | xAI | frontier | 6T MoE, training on Colossus 2 |