AI Briefing
KO

Open-weight models capture 56% of token share; OpenAI Astra's spend share is double that of Fable 5.1

·2026.09.17 09:00

Key point

Open-weight models captured 56% of token volume, marking the first time they exceeded a majority share, while OpenAI Astra secured a spend share more than double that of Fable 5.1.

1 / 2

Details

The Rise of Open-Weight Models and Cost Decline

As of August 2026, open-weight models accounted for 56% of total token volume, surpassing a majority share for the first time. This represents a sharp increase from less than 7% in December 2025. The increased adoption of open-weight models drove the average gateway token price down by 23.2% in August, marking the third consecutive month of decline and the largest drop since April.

Share Shifts for Anthropic and Google

Anthropic maintained 64% of total spend, but model-level shares fluctuated significantly. Fable 5's spend share plummeted from 13.2% in July to 4.9% in August; 90% of teams using Fable reduced their usage, with most migrating to Opus 5 (22.5% spend share). The judgment was that Fable's additional performance did not justify its double price.

Google's gateway token volume share dropped sharply from 30% to 5% due to the underperformance of Gemini 3 Flash. Half of the volume lost from Gemini 3 Flash shifted to lower-cost models like GPT-5.6 Luna, while most of the remainder distributed to higher-cost models such as Claude Opus 5 (approximately 9x price) and Sonnet 5 (approximately 3x price).

Early Performance of OpenAI Astra and Competitive Landscape

Launched on September 3, OpenAI Astra (GPT-6 Astra) showed strong initial performance, capturing 1/3 of OpenAI's total spend within 48 hours of launch. It held 7.7% of total gateway spend over 12 days, more than double the 3.7% share of Fable 5.1, which launched at the same price just two days later. Astra drove premium spend, proving that OpenAI can compete with other frontier labs in terms of scale and profitability.

Multimodal and Other Trends

In image generation, Google Nano Banana ranked first in spend (50%), while GPT Image ranked first in generation volume (46%). In video generation, Seedance ranked first in both generation volume and spend, while Google Veo's spend share rose to 20%. Conversely, xAI Grok Imagine's share of video generation dropped sharply from 42% in June to 19% in August.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.