Gemini 3.8 Flash hits 90.8% on Terminal-Bench at $0.75/$3.75
Cheap, fast coding at Flash prices — but the rate doubles Jan 1. Plus gated Flash Cyber, Meta's low-cost Muse Spark 1.3, and Anthropic's commerce agents.
Copy markdown
90.8% Terminal-Bench, at Flash prices
Gemini 3.8 Flash is GA on the Gemini API and now live across providers, scoring 90.8% on Terminal-Bench 2.1 (up from 81.6% for 3.7 Flash — the month's biggest jump) and 61.6% on SWE-Bench Pro. It's tuned for long-horizon coding and autonomous agents, with a 1M-token context at $0.75/$3.75 per 1M tokens.
The price doubles on January 1
That $0.75 input / $3.75 output rate is intro pricing through Dec 31, 2026; on Jan 1, 2027 it doubles to $1.50/$7.50. If you're sizing a Flash-based pipeline for next year, budget against the higher number now, not the launch teaser.
Flash Cyber ships gated
A separate Gemini 3.8 Flash Cyber variant — over 70% on internal vulnerability benchmarks and 47.2% pass@1 on CWE-Bench patching — is restricted to vetted defenders via Google's Fairwind Program, not general release. It's the newest general-plus-gated-cyber split, after OpenAI's Astra and Anthropic's Mythos.
Elsewhere: Meta's Muse Spark 1.3 undercuts the field
Meta put Muse Spark 1.3, plus a cheaper Contributor tier, on OpenRouter as one of the lowest-priced top-tier models — reportedly ~20% fewer tool calls and ~25% fewer tokens per task than 1.2. Cheaper agent loops if you route through it.
Elsewhere: Anthropic's commerce-agent blueprints
Anthropic published a Commerce Blueprint with reference shopping and merchant agents across retail, travel, telecom, and ticketing — fork-ready patterns if you're building a buying or checkout agent this weekend.