Google's Gemini 4 is in posttraining with internal benchmarks reportedly strong — 90%+ on MMLU-Pro, 85%+ on SWE-bench Verified. Expected release Q1 2027 (WSJ, 2026). The release is Google's response to losing enterprise AI share to OpenAI and Anthropic in 2025-2026.
Data last verified September 9, 2026 from WSJ, Google DeepMind, and investor materials.
Why the delay
Google's AI release cadence has slipped since the Gemini 3 launch (November 2024). Originally planned for mid-2026, Gemini 3.5 Pro was dropped because candidates weren't enough of an improvement over Flash. Gemini 4 then became the consolidated release (WSJ, 2026).
The 6-month delay reflects:
- Posttraining duration — RLHF + RLAIF + constitutional AI takes 4-6 months.
- Safety evaluations — UK AISI, US AISI, EU AI Act compliance testing.
- Multimodal integration — text + vision + audio + video + code in one model is harder to balance.
Internal benchmarks (per WSJ)
| Benchmark | Gemini 4 | Claude Opus 4.5 | GPT-6 Astra |
|---|---|---|---|
| MMLU-Pro | 91.2% | 88.7% | 90.1% |
| SWE-bench Verified | 85.4% | 74.1% | 76.8% |
| GPQA Diamond | 74.8% | 71.2% | 73.5% |
| MathVista | 82.1% | 79.6% | 80.7% |
| VideoMME | 78.3% | 75.1% | 77.2% |
Source: WSJ citing Google internal benchmarks (2026).
What about Gemini 3.8 Flash first
Google shipped Gemini 3.8 Flash on September 2, 2026 (see our coverage). Flash is the immediate commercial play; Gemini 4 is the technical flagship. The two releases fill the gap that Gemini 3.5 Pro left open (WSJ, 2026).
Strategic context
Google's AI strategy in 2026-2027:






