Gemini 3.5 Pro delayed for retraining

// 2h agoNEWS

Gemini 3.5 Pro delayed for retraining

Reports of a Gemini 3.5 Pro benchmark leak indicate that Google has postponed the model's release to July 17. The delay is reportedly intended to allow retraining on a stronger architecture rather than shipping a minor upgrade, with early results suggesting a major performance jump.

// ANALYSIS

Google is choosing to delay the release to retrain the model on a stronger architecture, prioritizing a major leap over an incremental upgrade.

* Delaying to July 17 indicates Google wants to ensure the model makes a significant impact upon release.

* Upgrading the underlying architecture suggests standard tuning was insufficient for their target benchmarks.

* The early benchmark leak signals a major jump in capabilities that could challenge other leading frontier models.

// TAGS

gemini-3.5-progoogleaibenchmarkleak

DISCOVERED

2h ago

2026-07-14

PUBLISHED

2h ago

2026-07-14

RELEVANCE

8/ 10

AUTHOR

adckerX

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

MODEL1h ago

OpenAI GPT-5.6 hits Amazon Bedrock

OpenAI's GPT-5.6 model family—including Sol, Terra, and Luna—is now generally available on Amazon Bedrock. Running on Bedrock's next-generation inference engine, the models support prompt caching with a 90% discount and match OpenAI's first-party pricing.

UPDATE2h ago

OpenRouter splits rankings by model weight

OpenRouter has updated its rankings platform by introducing separate leaderboards for open-weight and closed-weight models. This allows developers to track and compare usage statistics of proprietary, API-exclusive models against downloadable open-weight models.

UPDATE2h ago

Codex and Claude Code introduce advanced in-app browser capabilities, including multi-tab support and cookie imports, accelerating the shift toward autonomous computer use.

Codex has updated its in-app browser to support multiple tabs, cookie importing, and password persistence, with Anthropic's Claude Code quickly following with similar web-browsing capabilities. These upgrades allow AI agents to navigate authenticated sites and perform browser-based tasks alongside code editors and terminals. By embedding robust browser control directly into the agentic environment, developers can execute end-to-end workflows without leaving the command line or workspace app.