Up to 30% lower costs with up to 22% higher cache hits.
30% cost reduction on multi-turn coding and 17% on reasoning thanks to higher cache hit rates.
40% fewer tokens on file edits and up to 8% higher task completion on SWE and research evals.
Automatic context compaction at ~135k tokens.
Files API: Upload, list and download files from sandbox easily.
Credentials API for MCP servers, OAuth2, and 3rd‑party APIs as secure egress proxy.
Register secrets once with credentials.create() supporting bearer_token, environment_variable, or oauth2 (with auto‑refresh and token rotation).
Zero plaintext exposure: Secrets never enter model context, stdout, memory.
Egress proxy swaps real token on the request only for allow‑listed trusted_domains; any exfiltration attempt gets a 403.
Save reusable agent configs once with agents.create() and call them by id.
In early September 2026 I posted a series of announcements on X about the latest Gemini frontier models, pricing, API extensions, and benchmark breakthroughs.
Gemini 4 Argon is rolling out to a set of trusted cyber defenders with an output token limit of 1 M (up from 64 K).
Launch pricing for Gemini 4 Argon is $2 in / $10 out.
The Interactions API is now available on the Gemini Enterprise Agent Platform using the same google-genai SDK.
The new Credentials API injects secrets only for whitelisted domains, preventing sandboxed code from reading raw tokens.
Gemini 3.8 Flash TTS is free in Cline and scores 41 on the Artificial Analysis Intelligence Index, with 291 tok/s throughput and a 1 M context window.
Gemini 3.8 Flash ranks #1 on Hume’s Voice Design Benchmark and tops Voice Arena in six languages.
Furniture Assembly Benchmark top score rose from 28 % to 80 % in ten months.
Antigravity‑preview‑09‑2026 tools run natively on Gemini 3.8 Flash, bringing advanced sandboxing and context compaction.