Wednesday, September 2, 2026
Tech Beat
Sep 2, 2026, 6:13 PMArtificial Intelligence

Google Launches Gemini 3.8 Flash With DeepSWE Lead and Cyber Model

Google launches Gemini 3.8 Flash and Cyber, touting stronger coding, discounted API pricing and security gains, while computer control still trails Claude.

Listen to this briefingAudio briefing

Summary

Google released Gemini 3.8 Flash on September 2, 2026, its third Flash model in six weeks, while no frontier Gemini Pro model has arrived since early 2026 and the reportedly delayed Gemini 3.5 Pro remains absent. Google calls 3.8 Flash its best reasoning and coding model yet. Standard Flash targets agentic tasks and software development, while Flash Cyber shares its foundation but is tuned to detect and mitigate vulnerabilities. API access costs $0.75 per million input tokens and $3.75 per million output tokens through 2026, then rises to $1.50 and $7.50, as rival labs also cut prices to retain business users.

Google's benchmarks show marginal gains over Gemini 3.7 Flash on most tests but larger coding improvements. Gemini 3.8 Flash leads the DeepSWE software engineering leaderboard at its discounted cost. Its OSWorld 2.0 computer-use score improved but remains far behind Claude Opus, while GPT also performs poorly. Flash Cyber replaces version 3.5; Google says internal tests found more vulnerabilities and produced working patches more often. Chrome measured a 2.6 times increase in patch accuracy, and Google Cloud says the model found a critical vulnerability in two hours, with Wiz and Palo Alto Networks providing endorsements. Standard Flash is available across Google's ecosystem, requiring Pro or Ultra in the Gemini app, with free access through AI Studio. Cyber remains limited to trusted testers and governments.

Positives

  • Gemini 3.8 Flash leads the DeepSWE leaderboard for complex software engineering while using Google's discounted API rate.
  • Chrome measured a 2.6 times increase in patch accuracy with Gemini 3.8 Flash Cyber.
  • Google Cloud says Flash Cyber identified a critical vulnerability in two hours.
  • API pricing stays at $0.75 per million input tokens and $3.75 per million output tokens through 2026.
  • AI Studio offers free experimentation with Gemini 3.8 Flash, while paid Gemini app access requires Pro or Ultra.

Risks & concerns

  • OSWorld 2.0 results leave Gemini 3.8 Flash far behind Claude Opus in agentic computer use.
  • Regular API prices will double to $1.50 per million input tokens and $7.50 per million output tokens.
  • Gemini 3.5 Pro remains absent after a reported delay tied to coding performance.
  • Gemini 3.8 Flash Cyber is restricted to trusted testers and governments.
  • Performance evidence largely comes from Google benchmarks, internal tests, and endorsements from Wiz and Palo Alto Networks.
Primary sourceAI - Ars Technicahttps://arstechnica.com/ai/2026/09/google-releases-gemini-3-8-flash-its-third-flash-model-in-six-weeks/
Read full article
Editorial note: Tech Beat summarizes and analyzes third-party reporting. The source link is the authoritative article. This page does not reproduce the full source text.

More From The Wire

Artificial IntelligenceSep 2

TechCrunch Disrupt 2026 Adds Real World AI Stage With Nvidia, Robots and De-Extinction

Artificial IntelligenceSep 2

Meta Muse Voice Transcribe Brings 20+ Speaker Diarization for $0.18 an Hour

Artificial IntelligenceSep 2

OpenAI Astra’s Opaque Recurrence Alarms AI Safety Experts