Gemini 3.8 Flash (Vertex AI)

Google has shipped three Flash releases in about six weeks — Gemini 3.6 Flash on 21 July 2026, 3.7 Flash on 13 August 2026, and 3.8 Flash on 2 September 2026 — with Google calling 3.8 Flash its "most intelligent Flash model". The published spec line is identical to the previous Flash model, with the same 1,048,576 inpu Gemini 3.8 Flash adds low, medium and high thinking levels (up from minimal/low/medium/high on 3.6 and the low/medium/high set on 3.7) and Google benchmarks it against Claude Opus 5, Claude Sonnet 5 and the GPT-5.6 line rather than the newest OpenAI flagship. ComputingForGeeks ran calls against gemini-3.8-flash in Sept

Gemini 3.8 Flash is GA as the direct successor to Gemini 3.7 Flash, published with the model ID gemini-3.8-flash and a one-million-token context window (1,048,576 input tokens) with a maximum output of 65,536 tokens. Input modalities are text, image, video, audio and PDF, with text output, and supported tools include c Google positions the model as a meaningful upgrade for long-horizon coding, tool-using agents, finance and legal workflows, and scientific research tasks, while keeping the temporary $0.75 per 1M input and $3.75 per 1M output token prices. Kingy AI's six-call low-thinking API smoke check found that at high effort the m

Google released Gemini 3.8 Flash on September 2, 2026, marking its third Flash-tier model launch in just six weeks. The release comes in two variants: a standard "workhorse" Flash model suited to agentic tasks and software development, and Gemini 3.8 Flash Cyber, which is tuned for vulnerability detection and mitigatio For developers, API access is priced at an introductory rate of $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026, doubling to $1.50/$7.50 thereafter. Google published benchmark numbers showing marginal gains over Gemini 3.7 Flash on most tests but larger improvements in coding

Google released Gemini 3.8 Flash on September 2, 2026, three weeks after Gemini 3.7 Flash and one day after Anthropic shipped Claude Fable 5.1, with the pitch of frontier-class results on coding, finance, and legal agent benchmarks at a fraction of rival prices. Per Google's official model card, the model is built for Specifications include a 1-million-token context window, 64,000-token maximum output, multimodal inputs (text, images, audio, video), and customizable effort levels that trade quality against cost and latency. Knowledge cutoffs are mixed at March 2026 for some domains and January 2025 for others, with introductory pric

Third-party benchmark aggregator BenchLM tracks Gemini 3.8 Flash as a proprietary Google DeepMind release that went live on September 2, 2026 with a 1-million-token context window. The model scores 78.4 out of 100 overall and ranks 6th out of 232 tracked models, with its strongest eligible category being Coding at rank Speed and pricing metrics from BenchLM show 327 tokens per second output throughput with a 10.75-second time-to-first-token. API pricing is reported at $0.75 per million input tokens and $3.75 per million output tokens, with cached input priced at $0.075 per million and a blended rate of about $2.25 per million. The da

Google DeepMind released Gemini 3.8 Flash on September 2, 2026, as its fourth Flash model in under four months, positioning it as the most intelligent Flash-tier model to date. Artificial Analysis reports that the model scores 59 on the Intelligence Index with high reasoning, up three points from Gemini 3.7 Flash (high The three-point intelligence gain is driven primarily by stronger performance on agentic evaluations, including τ³-Banking (tool use), Terminal-Bench v2.1 (coding) and GDPval-AA v2 (real-world tasks), with the largest improvement on τ³-Banking at plus 12 points to 45 percent. Gemini 3.8 Flash inherits Gemini 3.7 Flash'

Google released Gemini 3.8 Flash on September 2, 2026 as its most intelligent Flash-tier model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows, sitting below the Pro tier while priced at a fraction of the cost. DataCamp reports it scoring 90.8 percent on Terminal-B Pricing holds at $0.75 input and $3.75 output per 1M tokens through December 31, 2026, then reverts to $1.50 and $7.50. Alongside the base model, Google introduced Gemini 3.8 Flash Cyber, a variant tuned for cybersecurity work with a real-world vulnerability discovery rate above 70 percent and a position on the CWE-Ben