Grok 5 (Upcoming)
Status
Expected: Q2-Q3 2026 (slipped from original Q1 2026 target) Polymarket: ~33% chance by June 30, 2026 Training: Active on Colossus 2 (MACROHARD)> ⚠️ Slipped timeline: Musk originally targeted Q1 2026, then Q2. As of mid-June 2026, no release. Likely Q3 for full API access.
What We Know
Parameters
- 6 trillion (MoE – only subset activates per query)
- xAI training a 10 trillion model in parallel (separate variant)[xai-10t-model-nextbigfuture.md]
Architecture
- Mixture-of-Experts (MoE) – roughly double Grok 4's estimated 3T
- Multi-agent system: expected expansion from 4 agents (Grok 4.20) to 16+ agents
- Native multimodal: text, images, audio, real-time video, temporal reasoning
- Context window: reported 1.5M tokens
Training Infrastructure
- Colossus 2: 550,000+ NVIDIA GB200/GB300 GPUs
- 400+ MW power, ~$18B hardware cost alone
- 7 models in training simultaneously (as of Apr 2026)[xai-10t-model-nextbigfuture.md]:
Features & Rumors
| Claim | Status | Source |
|---|
| -------- | -------- |
|---|
| Unsubstantiated (Musk claim) | Musk on X |
|---|
| Speculative blogs only | — |
|---|
| Logical extension of 4.20 | Analyst estimate |
|---|
| Mirrors Grok 4.3 pattern | Analyst estimate |
|---|
| Date | Significance for Grok 5 |
|---|
| ------ | ------------------------ |
|---|
| Feb 17, 2026 | Multi-agent architecture test |
|---|
| Feb 17, 2026 | Scaled multi-agent test |
|---|
| Apr 30, 2026 | 1M context, native video |
|---|
| Jun 4, 2026 | Voice interface |
|---|
| Jun 3, 2026 | #1 image-to-video |
|---|
| mid-Jun 2026 | 1.5T coding model |
|---|
| Compute Cost | Duration |
|---|
| ------------- | ---------- |
|---|
| ~$1B+ est. | ~2 months |
|---|
| ~$1.5B+ est. |
|---|
See Also
- concepts/grok-models – Full Grok model family
- concepts/grok-v9-medium – Coding-focused 1.5T model
