Google's fast, low-cost Gemini Flash model with strong agentic coding results, plus a Cyber variant for defensive security.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Gemini 3.8 Flash is a Flash-tier model in Google's Gemini family, positioned as a fast and comparatively inexpensive option for coding and agentic work. The official page text provided does not describe it directly, so the details below come from creator reports and may be incomplete.
Creators report about 73.7% on a long-horizon software engineering benchmark (DeepSWE), roughly tied with Claude Opus 5, at a much lower cost per task. They also report first place on Humanity's Last Exam and Terminal Bench 2.1, and strong results on finance, legal and agentic terminal tasks. It is reported as only average on GDPval and Terminal Bench 4.0, ranks around 10th overall on Artificial Analysis and lower on LiveBench, and uses many output tokens.
Reported API pricing is $0.75 input / $3.75 output per million tokens as an introductory rate, rising to $1.50 / $7.50. A Cyber variant is described as targeting defensive security and vulnerability patching. Creators used it through tools such as Cursor for demos including websites, 3D scenes and game clones, with mixed to good results. Official pricing, context window and availability details were not in the provided material.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Fast Flash model that beats larger models on finance, legal and agentic terminal coding, and ranks first on Deep SWE. Uses many output tokens and ranks lower on LiveBench; a Cyber variant targets defensive security and vulnerability patching.
Cheap, fast Google model at $0.75/$3.75 per million tokens, scoring 73.7% on the coding benchmark at far lower cost per task than Opus 5 or Fable. Its Megabonk clone looked decent and was generated via Cursor.
Scored 73.7% on the Deep Sweep coding benchmark, narrowly beating GPT-6 Astra. The creator said this makes him doubt the benchmark.
New Google Gemini Flash model priced at $0.75/$3.75 per million tokens (introductory, rising to $1.50/$7.50). It scores 73.7% on the long-horizon software engineering benchmark, first on Humanity's Last Exam and Terminal Bench 2.1, but is only average on GDPval and Terminal Bench 4.0. The creator's demos (3D biomes, product websites, a PowerPoint deck, an Everest topographic map, a Doom clone) were mixed to good, and he calls it not quite frontier but very competitive on price.
Cyber-focused variant of Gemini 3.8 Flash with fewer cyber guardrails, available only to trusted defenders through a program. It scores 86.2 on CyberGym, ahead of GPT 5.5 Cyber at 85.6, and Google also reported a large gain over 3.7 Flash on an internal 20-language vulnerability discovery benchmark.
Released the same day and ranked number one on a software engineering leaderboard; creator will cover it later.
Cheap, fast coding-strong model scoring about 74% on DeepSWE, tied with Claude Opus 5 and above Fable 5, at $2.36 per task versus much higher for those. Creator considers it underhyped; it ranks only around 10th on Artificial Analysis overall and uses many output tokens.
Closest neighbours by meaning, not just by category.
Model · Google
Google's faster, lower-cost Flash model in the Gemini 3.5 family, aimed at agentic and coding workloads.
Model · Google
Google's fast, lower-cost multimodal Flash-tier Gemini model for the Gemini app, AI Studio and coding tools.
Model · Google
Google's Gemini Flash-tier model update, reported to use fewer tokens and cost less than Gemini 3.5 Flash.
Model · Google
Google's rumored fast, low-cost Gemini model, reportedly positioned below Pro; not confirmed by official pages.
Model · Google DeepMind
Google DeepMind's lightweight, low-cost Gemini 3.5 model aimed at fast, high-volume tasks.
Model · Google DeepMind
Variant of Google DeepMind's Gemini Flash model family reportedly tuned for cybersecurity tasks.
