Bitget App
Trade smarter
Buy cryptoMarketsTradeFuturesStocksEarnInstitutionAI & More
On the eve of Anthropic's IPO, OpenAI launches its strongest model yet, GPT-6.2 Astra, claiming performance surpasses Fable 5.2

On the eve of Anthropic's IPO, OpenAI launches its strongest model yet, GPT-6.2 Astra, claiming performance surpasses Fable 5.2

华尔街见闻华尔街见闻2026/11/17 20:33
Show original
By:华尔街见闻

On the eve of Anthropic's IPO, OpenAI made a "precise strike" by launching the dual models GPT-6.2 Astra and Sol. GPT-6.2 Astra is designed for complex professional scenarios, covering areas such as financial modeling and scientific research. In multi-step professional benchmarking tests, Astra scores 5.5 points higher than Claude Fable 5.2, while costing only 5% of the latter. GPT-6.2 Sol is oriented toward daily tasks, with costs further substantially reduced compared to Astra.

OpenAI has launched the latest iteration of its flagship model, which not only improves performance but also significantly reduces usage costs, surpassing leading competitor Anthropic's top model in multiple benchmark tests.

The newly released GPT-6.2 Astra is tailored for complex professional scenarios, covering areas such as financial modeling and scientific research, while the simultaneously launched GPT-6.2 Sol is designed for everyday tasks.

OpenAI stated that GPT-6.2 Astra scored 5.5 points higher than the competing Claude Fable 5.2 in multi-step professional task benchmarks, while costing only 5% of the latter. The launch coincides with Anthropic’s much-anticipated pre-IPO period, making the timing quite sensitive.

On the eve of Anthropic's IPO, OpenAI launches its strongest model yet, GPT-6.2 Astra, claiming performance surpasses Fable 5.2 image 0

The introduction of the new model comes amid ongoing external scrutiny of safety risks associated with advanced AI systems. OpenAI previously suspended the release of GPT-6.1 Astra at the end of September, citing the model's “failure to adequately restrict scope and permission boundaries.” The company had also temporarily halted training on its most advanced AI systems, only resuming after cutting internet access to the models.

Safety Concerns Led to Nearly Two-Month Delay in Release

The release of GPT-6.2 Astra comes nearly two months after the previous version was halted. At the end of September, OpenAI indicated that GPT-6.1 Astra did not meet expectations regarding “defining work scopes and authorization boundaries, and providing feedback to users about completed work types,” thus the release was postponed.

Previously, OpenAI also briefly paused training on its most advanced AI systems. This decision was triggered by a series of safety incidents—test models had accessed third-party online platforms without authorization. Training resumed only after OpenAI cut off the affected model’s internet access.

Anthropic faces similar safety challenges. Both leading AI startups have expressed that given a series of security incidents, slowing development is a reasonable step.

Performance Improvement and 20% API Price Reduction

From a technical standpoint, GPT-6.2 Astra set new records in several key benchmark tests. In the “Agents' Last Exam” multi-step professional task benchmark, the model scored 5.2 points higher than its predecessor, GPT-6 Astra, while costing less than one-third of the former.

In the scientific task benchmark “Terminal-Bench-Science 0.1,” its score was also 9 percentage points higher than the previous generation, with costs still less than one-third.

On the eve of Anthropic's IPO, OpenAI launches its strongest model yet, GPT-6.2 Astra, claiming performance surpasses Fable 5.2 image 1

In terms of API pricing, input and output token prices for GPT-6.2 Astra are 20% lower than those of GPT-6 Astra, and cached input enjoys a 60% discount. The price per million tokens is: $8 for input, $0.40 for cached input, and $40 for output.

The simultaneously released GPT-6.2 Sol is positioned as a model for everyday tasks, with API token prices remaining consistent with GPT-6.1 Sol.

In the “FrontierCode 2.0” code-writing benchmark, the score of GPT-6.2 Sol was within 0.5 points of GPT-6.2 Astra, with costs at about one-fifth of the latter.

On the eve of Anthropic's IPO, OpenAI launches its strongest model yet, GPT-6.2 Astra, claiming performance surpasses Fable 5.2 image 2

In the “LifeVal” daily task evaluation benchmark, GPT-6.2 Sol scored higher than GPT-6 Astra while costing about one-seventh as much as Astra.

On the eve of Anthropic's IPO, OpenAI launches its strongest model yet, GPT-6.2 Astra, claiming performance surpasses Fable 5.2 image 3

Professional Documentation Output Closer to Industry Standards

OpenAI has placed significant emphasis on professional document quality in this release.

The company claims that slides, spreadsheets, and documents generated by GPT-6.2 Astra follow user instructions and industry practices more rigorously, with a noticeably reduced “AI feel.” The writing style is more direct and specific, with fewer generic summaries and redundant content.

In the “Harvey LAB-AA v1.1” legal document evaluation benchmark, GPT-6.2 Astra’s pass rate was more than double that of Claude Fable 5.2, while costing about one-sixth as much.

On the eve of Anthropic's IPO, OpenAI launches its strongest model yet, GPT-6.2 Astra, claiming performance surpasses Fable 5.2 image 4

In OpenAI’s internal “GDPval v2” benchmark, which covers real-world projects in fields such as finance, data science, biotechnology, and marketing, GPT-6.2 Astra improved by 6.2 percentage points over its predecessor, while reducing output tokens by 22%. GPT-6.2 Sol showed a 5.3 percentage point increase over GPT-6.1 Sol.

On the eve of Anthropic's IPO, OpenAI launches its strongest model yet, GPT-6.2 Astra, claiming performance surpasses Fable 5.2 image 5

Valuation and Listing Competition Under Competitive Pressure

The timing of this release is clearly competitive. OpenAI chose to launch its new model just before Anthropic’s public debut on Wall Street, as both companies are fiercely contending for enterprise clients.

At the same time, OpenAI itself is at a crucial stage of capital operations. Reportedly, the company is currently undergoing a new round of financing with a valuation of $1.25 trillion, and has delayed its IPO plans until next year.

As the financing window nears, offering stronger performance at lower costs may become a key factor for OpenAI to demonstrate its commercial competitiveness to potential institutional investors.

0
0

Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.

Understand the market, then trade.
Bitget offers one-stop trading for cryptocurrencies, stocks, and gold.
Trade now!

You may also like

Nvidia's "Hidden" Key Market: Singapore Subsidiary Reportedly Exceeded $100 Billion in Revenue Last Year, Up 72% Year-on-Year

According to invoice address statistics, in Nvidia's fiscal year 2025, Singapore was the company's second largest source of revenue after the United States. Starting from the third quarter of fiscal year 2026, which ends in October 2025, Nvidia adjusted its regional revenue disclosure method and no longer separately lists Singapore’s revenue. It is reported that Nvidia’s Singapore subsidiary achieved a revenue of $104 billion in fiscal year 2025, a year-on-year increase of 72%. Although the subsidiary's revenue exceeded $100 billion, its profit was only $53 million.

华尔街见闻•2026/11/17 20:03

Nike Announces Five-Year Recovery Plan: Aiming to Return to Industry Growth Rate by Fiscal Year 2029

After nearly a decade, Nike once again held its Investor Day, attempting to explain to the market how it plans to emerge from years of prolonged operational downturn. On November 17, Nike in the US...

华尔街见闻•2026/11/17 20:02