Two Speed Records Today: Gemini 3.7 Flash GA for Coding Agents, GPT-5.6 Sol Ultrafast at 750 Tok/s

Two new top-tier inference options landed today: Gemini 3.7 Flash is GA for coding agents at $0.75/1M input tokens; GPT-5.6 Sol Ultrafast previews at 14× speed (750 tokens/sec) via Cerebras.

Gemini 3.7 Flash (GA, August 13) is Google's new workhorse model for coding and agentic workloads, available now in the Gemini API and AI Studio. The headline improvements: better issue resolution, fewer failed agent loops, and higher-fidelity code generated directly from design mocks. Introductory pricing: $0.75/1M input, $3.75/1M output through December 31, 2026.

The standout feature is tunable reasoning per request via thinking_level in generation_config:

• Low — fast, lowest cost; tool calls, lookups, simple steps
• Medium (default) — recommended for complex code and multi-step agents
• High — maximum reasoning; hard problems, complex debugging

GPT-5.6 Sol Ultrafast (limited preview): OpenAI is previewing a Cerebras-powered tier that runs GPT-5.6 Sol at up to 750 output tokens/second — roughly 14× faster than Standard processing. Available now to a select group of API customers; targets real-time and highly interactive applications.

Why it matters: Per-request thinking-level routing inside a single model endpoint is a cleaner pattern than managing multiple models. And 750 tokens/second changes what "real-time agent UI" means — streaming a full code diff in under two seconds becomes viable.

Sources: Gemini 3.7 Flash — blog.google · Gemini 3.7 Flash API docs · Previewing Ultrafast — openai.com