Loading summary...
5 min read (71% time saved)
OpenAI is strongly suspected to be on the verge of releasing a new model, GPT-5.1, with evidence suggesting it is the recently appeared "Polaris Alpha" model on Open Router, supported by JavaScript bundle leaks, Polaris Alpha's unique performance and generation style, its exclusive appearance on Open Router, and recent infrastructure changes within OpenAI.
Summarize another video
Press ⌘K to quickly paste a new URL
The video introduces the strong belief that OpenAI is about to release a new model, GPT-5.1, distinct from the recently dropped Codeex Mini. This speculation is fueled by leaked JavaScript bundles and the emergence of a new, high-performing model called Polaris Alpha on Open Router. The speaker, despite lacking direct insider information, feels confident about the impending release and its potential impact.
This section is a sponsored message for WorkOS, highlighting its role in providing enterprise authentication solutions like SSO for major tech companies such as OpenAI, Anthropic, Snowflake, and Carta. WorkOS simplifies complex authentication requirements, saving companies from dedicating significant engineering resources to building and maintaining these systems internally. It offers various features like directory sync, vault, radar, and MCP auth, with a generous free tier for the first million users.
Further evidence for GPT-5.1's release comes from JavaScript bundles found on OpenAI's static CDN. These bundles explicitly contain references to "GPT51 Pro," "GPT51," and "GPT51 Reasoning." The speaker emphasizes that such details within bundled code are difficult to hide and strongly suggest an upcoming launch.
The speaker posits that Polaris Alpha is GPT-5.1, noting its impressive throughput of 114 TPS, significantly faster than GPT-5's average of 41 TPS. While latency is similar, the speed is uncharacteristic of current OpenAI models. More tellingly, Polaris Alpha's image generation style, with specific UI elements, gradients, and glowing effects, closely mirrors the aesthetic often produced by GPT-5 models, which is distinct from other models like Opus.
The video discusses the pattern of anonymous model drops on Open Router. Historically, models that appear only on Open Router, without being simultaneously released elsewhere, have consistently been identified as OpenAI models (e.g., Horizon Alpha and Polaris Alpha). This exclusivity on Open Router is presented as a key indicator that Polaris Alpha originates from OpenAI, differentiating it from models from other labs that tend to distribute more broadly.
Sam Pake's EQbench, a writing benchmark, shows Polaris Alpha clustering closely with other OpenAI models, including the Horizon series, in terms of its data and behavior. The speaker performs a "vibe check" by prompting Polaris Alpha to defend Java, noting its tendency to generate "endless lists"—a characteristic behavior frequently observed in GPT-5 models. The speaker recommends T3 Chat for trying Polaris Alpha and other stealth models.
Recent updates to OpenAI's Codeex model are highlighted as indirect evidence. Codeex received a "small upgrade," becoming smarter, more collaborative, and 3% more token efficient. Concurrently, OpenAI increased rate limits for ChatGPT Plus business and educational accounts by 50%, attributing this to "efficiency improvements to get more out of our GPUs." The speaker interprets this as OpenAI freeing up significant GPU resources, likely for the deployment or training of a new, larger model like GPT-5.1.
OpenAI models uniquely incorporate "developer messages" as a layer between system prompts and user messages. When directly asked, Polaris Alpha acknowledges the existence of developer messages. Furthermore, a jailbreak method involving fake system and developer tags causes Polaris Alpha to reveal its internal background, stating it is "a large language model from an unknown provider," which, in context, strongly suggests OpenAI's involvement in its development and stealth release.
The speaker concludes with high confidence that Polaris Alpha is indeed GPT-5.1, despite its unexpectedly high TPS. The cumulative evidence—including JavaScript leaks, the model's generation style, its exclusive presence on Open Router, Codeex infrastructure shifts, and the developer message/jailbreak findings—all point towards OpenAI. The speaker expresses excitement for further interaction with the model and invites viewer opinions.
"I do really think 5.1 is about to happen." — [The video's host]
"There's very little that you can hide once it's in the bundle." — [The video's host]
"Every time a model only dropped on Open Router and nowhere else, that model ended up being OpenAI." — [The video's host]
"It seems like a lot of focus went into the efficiency here because they needed more GPUs for something." — [The video's host]
"I'm relatively confident that we just got our first taste of GPT 5.1." — [The video's host]
Theo - t3․gg
7 min read
WorldofAI
5 min read
The PrimeTime
4 min read
No Priors: AI, Machine Learning, Tech, & Startups
8 min read
Theo - t3․gg
4 min read
Rate Limited
8 min read
