जेमिनी 3.8 Flash

Google's latest workhorse Flash model, delivering its predecessor's speed and price with a marked step up in software engineering, agentic work and multi-step reasoning in specialised domains. It accepts text, images, audio, video and files across a million-token context, and carries three thinking levels. Google is explicit that it works harder than 3.7 Flash -- more reasoning steps and more iterative tool calls -- so it can spend more tokens on the same task, especially at higher effort.

This is the third Flash release in six weeks, and the pattern holding across all three is that the price does not move. Google calls 3.8 Flash its most intelligent workhorse model and puts it out at exactly what 3.7 Flash costs, aiming the gains at software engineering, agentic tasks and multi-step reasoning in specialised domains rather than at general chat.

Its reported results follow that aim. On DeepSWE v1.1, which measures long-horizon software engineering, Google says it outperforms most larger frontier models at autonomously solving complex engineering problems end to end, at a fraction of their cost. It reports 54.9% on HLE-Verified for multi-step reasoning across STEM, humanities and professional fields, and says it exceeds 3.7 Flash and other frontier models on the Vals Finance Agent V2 and Harvey Legal Agent benchmarks - two evaluations pointed at professional domains rather than at coding. Google also reports a significant leap in prompt-injection robustness on Gray Swan, which matters more than usual for a model this cheap to run in an agent loop.

The interesting caveat is one Google raises itself, and it is a cost caveat rather than a quality one. It says 3.8 Flash "works harder": it executes extra reasoning steps and calls tools iteratively, and it may therefore use more tokens to reach a better answer, especially at higher effort. Google's own recommendation is that if efficiency matters more than the last few points of quality, 3.7 Flash remains the model to reach for - the two are siblings on the same price sheet, not a replacement and a legacy.

The price sheet has a date on it, which is the one thing about this model most likely to catch someone out. The $0.75 input and $3.75 output rates run through 31 December 2026 and double to $1.50 and $7.50 on 1 January 2027, with context caching moving from $0.075 to $0.15 in step. That is introductory pricing, not a permanent rate.

Google shipped a second model in the same post, Gemini 3.8 Flash Cyber, its most capable cybersecurity model, aimed at vulnerability detection and automated patching and available to trusted defenders through a new Fairwind Program. That one is not served here; everything above describes the general model.