Google and Meta Debut AI Models on Same Day, Benchmark Results Divided

On Wednesday, Google and Meta released their newest frontier AI models just hours apart. Google launched Gemini 3.8 Flash alongside a cybersecurity version, while Meta introduced Muse Spark 1.3.

The two releases invite a direct comparison. Independent testing by Artificial Analysis yields split results. Meta leads in agentic knowledge work and scientific reasoning, while Google holds an edge in factual recall and terminal coding.

Two Frontier Releases Land on the Same Day

Gemini 3.8 Flash is Google's third Flash release within six weeks. The model is priced at $0.75 per 1 million input tokens and $3.75 per 1 million output tokens.

Today we’re introducing a new 3.8 Flash model, our 3rd Flash release in just 6 wks.

It delivers significant leaps from 3.7 Flash across software engineering, agentic tasks, and multi-step reasoning. On DeepSWE v1.1, it outperforms most larger frontier models in autonomously… pic.twitter.com/MB5seliluh

— Sundar Pichai (@sundarpichai) September 2, 2026

The introductory rates are valid until December 31, 2026. After that, prices double to $1.50 and $7.50 per 1 million tokens.

Google also paired the general model with Gemini 3.8 Flash Cyber. Access is restricted to a set of trusted defenders. On CyberGym, a benchmark for finding vulnerabilities, it scored 86.2%.

The cybersecurity model also achieved 47.2% on CWE-Bench, a patching benchmark. Google reported that the model produced 2.6 times more correct patches for Chrome vulnerabilities than larger commercial models.

Access is behind the Fairwind Program, which limits usage to government authorities and critical infrastructure operators. A day earlier, OpenAI drew a similar boundary around Astra, the first model it rated at a critical cybersecurity threshold.

Meanwhile, Meta rolled out Muse Spark 1.3 via Muse Code and the Meta Model API. Company engineers measured roughly 20% fewer tool calls than version 1.2.

Muse Spark 1.3 is rolling out today with frontier performance almost too cheap to meter. This is the biggest jump we've made so far on coding and agentic work. Try it in Muse Code and our API.

Next up 🍉 and Muse Spark open weights releases coming soon. pic.twitter.com/XQQEDEJGD7

— Mark Zuckerberg (@finkd) September 2, 2026

Gemini 3.8 Flash vs Muse Spark 1.3: Independent Benchmarks Split the Result

Artificial Analysis tested the models, and the results show how they rank. In max mode, Muse Spark 1.3 scored 1,754 Elo on GDPval-AA v2. Gemini 3.8 Flash (high) returned 1,545.

Meta also led the Sierra Research banking agent test, with 52.4% to 44.9%, and CritPt physics reasoning. Gemini 3.8 Flash led Terminal-Bench 2.1 at 87.6%, AA-LCR long context at 81%, and AA-Omniscience accuracy at 55%.

Gemini 3.8 Flash posted the highest GPQA Diamond score among tested models, at 95%. The two finished within one point of each other on Humanity's Last Exam.

Meta's top scorer does not ship today. The company said max reasoning will arrive after further safety testing, leaving xhigh as the available variant.

That version scored 61 on the Artificial Analysis Intelligence Index, four points above Muse Spark 1.2. It trails Claude Fable 5.1 at 66 and Claude Opus 5 at 63.

Additional releases are already queued. Elon Musk has said Grok 4.7 arrives shortly, which would mean four frontier launches within a fortnight.