Google delays Gemini 3.5 Pro release after disappointing coding performance in June training update
Google has delayed the release of Gemini 3.5 Pro past its June deadline due to coding performance issues. The company retrained the model in late June with new data but saw disappointing results, according to Bloomberg. An upgraded Flash model is now in testing with partners.
Google delays Gemini 3.5 Pro release after disappointing coding performance in June training update
Google has pushed back the release of Gemini 3.5 Pro past its June deadline after coding performance failed to meet expectations, according to Bloomberg.
What happened
Google announced Gemini 3.5 Flash at I/O 2026 in mid-May and said the Pro version would arrive in June. That deadline has passed with no update.
According to Bloomberg, Google is "taking time to try to improve [Gemini 3.5 Pro's] capabilities, particularly in coding." In late June, Google updated the training data in an attempt to improve coding skills, but the results were disappointing.
The timeline suggests development saw a reset between I/O and the missed launch. Performance in other domains remains unclear. Gemini 3.1 Pro, the current flagship model, dates back to February 2026.
What Google says
In a statement, Google said it is "currently testing 3.5 Pro, an upgraded Flash model, and other models with partners." The company added: "We're shipping quickly across a wide range of models while keeping them highly cost-effective for customers."
No new timeline for Gemini 3.5 Pro's release was provided.
Internal coding context
As of April 2026, 75% of all new code at Google is AI-generated and approved by engineers, up from 50% last fall, according to Bloomberg. However, efforts face resistance from engineers who believe important code should be human-written to adhere to Google standards, according to ex-employees.
Internally, engineers are facing AI capacity restraints with coding tools. Google is working to "unite the company's internal artificial intelligence coding tools."
Google DeepMind (AI Studio), Cloud (Vertex), and the Android team (Android Studio) each maintain separate AI coding tool efforts.
What this means
The delay signals Google is prioritizing quality over speed for its flagship Pro model, particularly in coding—a domain where OpenAI and Anthropic have set high benchmarks. The fact that a late June retraining attempt failed suggests deeper architectural or data challenges rather than simple fine-tuning issues. With 75% of Google's own code now AI-generated, the company faces pressure to ship coding models that match internal standards while competing externally.
Related Articles
Google Launches Gemini 3.5 Transcribe, a Speech-to-Text Model That Cleans Up Rambling Speech
Google has released Gemini 3.5 Transcribe, a new speech-to-text model that automatically detects over 85 languages, removes filler words, and structures unstructured speech into clean text. The model powers Android's Rambler feature and is rolling out to Chrome, Docs, Gmail, and other Google products.
Google Launches Gemini 3.5 Transcribe with 2.6% Word Error Rate, Powers Gboard Rambler
Google has released Gemini 3.5 Transcribe, a speech-to-text model claiming a 4.0% word error rate in streaming mode and 2.6% in non-streaming mode, according to benchmarks from Artificial Analysis. The model already powers Gboard Rambler on Android and the Gemini app for macOS, with Chrome support coming next.
Google Launches Gemini 3.5 Transcribe with 4.0% Word Error Rate Across 85 Languages
Google has released Gemini 3.5 Transcribe, a speech-to-text model that automatically detects 85 languages, removes filler words, and corrects misspoken phrases. The company claims a 4.0 percent word error rate for streaming audio and 70 percent lower latency than its predecessor, Chirp 3.
Google DeepMind Launches Gemini 3.5 Transcribe, Claims 2.6% Word Error Rate in Testing
Google DeepMind has released Gemini 3.5 Transcribe, a speech-to-text model available via two APIs for real-time streaming and pre-recorded audio. According to Artificial Analysis benchmarks cited by Google, the model achieves a 2.6% word error rate for non-streaming transcription and 4.0% for streaming.
Comments
Loading...