Google has released Gemini 3.7 Flash, its new coding and autonomous agent model, now available globally as a low-cost, general-purpose AI offering. At the same time, OpenAI introduced a limited preview of GPT-5.6 Sol Ultrafast, a new tier for its top performing model, powered by Cerebras hardware and designed for significantly faster output.
ETH
SOL
BSC
ROBINHOOD
USDT
AAPLGemini 3.7 Flash now broadly available
Gemini 3.7 Flash is a major update in Google’s suite of artificial intelligence models, with a focus on software engineering, web development, and advanced knowledge work. The model supports input sizes up to a million tokens—equivalent to roughly 750,000 words—and handles various formats including text, images, video, audio, and PDFs. It also features tool-calling abilities for controlling computers and managing complex workflows, aligning with Google’s push towards real-time, autonomous agents.
According to Google, Gemini 3.7 Flash can complete coding tasks in about 2 minutes and 13 seconds, less than half the time required by its previous Flash release. The company stated that the new model offers both improved quality and speed, aiming to provide a “workhorse” tool for developers and enterprise users needing rapid, high-volume AI output.
Pricing for Gemini 3.7 Flash is set at $0.75 per million input tokens and $3.75 per million output tokens through the end of this year—half the original cost for Gemini 3.6 Flash. These rates will double to $1.50 and $7.50, respectively, after December 31. Google’s benchmarks indicate that the new model leads competitors such as Claude Sonnet 5 and GPT-5.6 Terra across 11 of 18 test categories, including a top Elo score of 1,588 in Code Arena web development and 30.4% on the AutomationBench metric for enterprise workflow automation.
Mini dictionary: Claude Sonnet 5 is the latest mid-tier large language model from Anthropic, designed to balance cost, speed, and accuracy for enterprise AI applications.
Gemini 3.7 Flash is designed as a highly capable and efficient model, offering substantial gains across software engineering, web development, and complex knowledge work, while also maintaining affordability for developers worldwide.
| Model | Input Token Limit | Top Web Dev Elo | AutomationBench (%) | Introductory Price (per million input/output tokens) | Availability |
|---|---|---|---|---|---|
| Gemini 3.7 Flash | 1,000,000 | 1,588 | 30.4 | $0.75 / $3.75 | General access (160+ countries) |
| Claude Sonnet 5 | Not specified | Below Gemini 3.7 | Below Gemini 3.7 | Not specified | General access |
| GPT-5.6 Terra | Not specified | Below Gemini 3.7 | Below Gemini 3.7 | Not specified | General access |
| GPT-5.6 Sol Ultrafast | Not specified | Not specified | Not specified | Invite-only | Limited preview |
OpenAI’s speed leap with GPT-5.6 Sol Ultrafast
OpenAI, the company known for its GPT line of advanced language models, has unveiled GPT-5.6 Sol Ultrafast in a restricted preview. This new tier leverages specialized wafer-scale chips from Cerebras, a US-based AI hardware manufacturer, to achieve a throughput of up to 750 tokens—roughly 560 words—per second. The performance jump is intended to enable applications such as real-time voice agents and advanced autonomous business tools.
Mini dictionary: Cerebras is a company specializing in large wafer-scale processors, specifically designed for AI workloads, which can deliver higher throughput and faster processing compared to traditional GPU-based systems.
GPT-5.6 Sol Ultrafast is not a new model, but a speed-optimized offering based on the current GPT-5.6 Sol, a foundation model that was recently reinforced by AI red teaming against security holes such as prompt-injection attacks. The Ultrafast tier initially targets select customers via invite, with broader access planned as capacity increases.
Early feedback highlighted that Cerebras-powered performance enables new interactive experiences, for example, allowing a voice agent to process information and respond in near real time during ongoing phone calls.
Industry trends and model access
Both Google and OpenAI signaled a shift from headline-grabbing intelligence benchmarks to a focus on real-time agent capabilities. Google’s aggressive pricing and wide global access sets Gemini 3.7 Flash apart; the model is now live in more than 160 countries. In contrast, OpenAI’s GPT-5.6 Sol Ultrafast remains exclusive to selected users, pending a wider deployment.
The rollout of Gemini 3.7 Flash appears timed to address user demands for faster and more autonomous agents, especially as Google’s premium Gemini 3.5 Pro model has yet to see release. Meanwhile, OpenAI relies on third-party hardware for speed improvements rather than its own in-house chips, signaling a pragmatic approach to keeping up with market competition.
Industry observers note that as both companies race to deliver instant-response AI agents, access and affordability could become as important as raw model capability.
Gemini 3.7 Flash and GPT-5.6 Sol Ultrafast underline this new competitive dynamic, with Google making its latest AI engine widely accessible, while OpenAI tests the fastest version of its technology with a limited group of business users.
