Back to News
IndustryAI Understanding briefing

Google DeepMind confirms Gemini 4 has entered post-training phase

Google DeepMind’s new leadership has confirmed that the next-generation Gemini 4 model is in post-training, aiming for a release before the end of 2026.

4 min readRead the linked source
Source-provided image accompanying Google DeepMind confirms Gemini 4 has entered post-training phase
Source referenceSource recorded
Publisher
tech-insider.org
Source link
tech-insider.orghttps://tech-insider.org/gemini-4-google-launch-deepmind-post-training-2026/
Source type
Linked source — primary-source status has not been established.
ContextUnderstand this in 60 seconds

Start here

Key terms

Post-training
Training steps applied after pretraining, such as instruction tuning, preference optimization, and safety tuning.
Reinforcement Learning from Human Feedback (RLHF)
A training method that uses human preference signals to shape model behavior.
Artificial Intelligence (AI)
The broad field of building systems that perform tasks requiring pattern recognition, reasoning, language, or decision-making.
Test yourselfAI Models Explained Quiz

What happened

Google DeepMind has officially entered the phase for its upcoming flagship AI model, Gemini 4. In his first media interview since assuming leadership of the division, Koray Kavukcuoglu stated that the model is being refined for safety, instruction-following, and reliability, with a target release date set for 'much earlier' than the end of 2026. This development follows a period where Google has relied on incremental updates to its Gemini 3.8 Flash model while competitors have launched new generational flagships.

The confirmation of Gemini 4's status came via an interview with Koray Kavukcuoglu, the new head of Google DeepMind. is the final development stage where pre-trained model weights are tuned through reinforcement learning from human feedback (RLHF) and safety alignment to ensure the model is suitable for public deployment.

As of September 18, 2026, Gemini 4 was not listed in Google's official product catalog, confirming that the model is not yet available for public or developer use. The current flagship, Gemini 3.8 Flash, was released on September 2, 2026, as an incremental update to the existing Gemini 3 architecture.

The announcement arrives during a highly active month for the AI industry. Within the last three weeks, OpenAI released GPT-6 Astra and Anthropic launched both Claude Fable 5.1 and Claude Opus 5.5, effectively resetting the performance benchmarks for the industry.

Source details: tech-insider.org ↗

Why it matters

The transition to marks a critical step in Google's attempt to regain its competitive standing in the frontier AI market. Currently, independent benchmarks place Google's Gemini 3.8 Flash significantly behind rivals like OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5.5. Because enterprise and developer adoption is increasingly driven by performance metrics, Google’s current reliance on a 'budget-tier' model has created a strategic gap. A successful Gemini 4 launch is essential for Google to remain a top-tier contender in reasoning and agentic tasks, as the company's distribution advantages in Android and Workspace are currently being tested by the superior intelligence scores of competing models.

Google is currently facing a 'generational gap.' While the company pioneered the transformer architecture, its recent releases have been point-updates rather than the full-generation leaps seen from OpenAI and Anthropic. This has left Google without a 'frontier-leading' entry on major intelligence leaderboards.

Independent data from sources like the Artificial Intelligence Index shows Gemini 3.8 Flash trailing Claude Opus 5.5 by 17 points. This performance deficit is significant for enterprise procurement teams who prioritize reasoning capabilities for coding and agentic workflows.

The company's strategy of using Gemini as a cost-effective, high-throughput model is being challenged by Anthropic’s recent pricing shifts, which offer high-performance models at lower costs. Gemini 4 represents a necessary pivot to reclaim the 'premium' market segment.

Interactive Mechanism

Interactive Mechanism: How It Actually Works

Explore the underlying technology behind this development interactively.

Thinking Budget (Test-Time Tokens):1,024 tokens
Complex Accuracy79%Math & Code Logic
Latency3.2sTime to first full output
Inference Cost$0.0092Per query estimated
Reasoning StyleStep VerificationInternal chain depth
Active Thinking Trace:
1Deconstruct user problem into formal constraints
2Propose candidate hypotheses & step-by-step calculation
3Self-correction: Backtrack and refute subtle edge cases
4Exhaustive consistency check & final output synthesis
Core takeaway: Test-time compute fundamentally changes AI economics. Instead of only scaling during pre-training, giving reasoning models more tokens at inference time allows them to systematically solve PhD-level STEM problems.
Interactive Concept Check+10 Points
AI Models Explained Quiz

In AI, what are a model's "parameters"?

What to watch next

The primary uncertainty remains the specific release date and the model's performance capabilities. While Kavukcuoglu indicated a launch before year-end, no concrete timeline, benchmark data, or pricing structure has been disclosed. Observers should monitor whether Gemini 4 can close the 17-point intelligence index gap currently separating Google from Anthropic’s latest offerings. Additionally, it remains to be seen if Google will introduce multiple variants, such as a 'Pro' or 'Flash' version of the new generation, to compete across different price and performance tiers.

Watch for official announcements regarding the model's architecture and modality support. The current report lacks details on whether Gemini 4 will feature expanded context windows or new native multimodal capabilities.

Monitor the competitive response from OpenAI and Anthropic. As Google prepares its launch, competitors may further adjust their pricing or release additional 'Sol' or 'Luna' variants to maintain their current lead.

Developers should continue building on existing Gemini APIs, as Google has historically maintained backward compatibility during model transitions. There is no current evidence that the Gemini 4 launch will deprecate existing 3.8-series services.

Related guides & quizzes

AI Models ExplainedTransformersFuture of AIAI TrainingTest what you know — try a free AI quizLook up an AI term in our glossaryFollow the AI funding tracker
Found this useful?