Asana makes its browser agent 76x cheaper and 5x faster with GPT-6.1 Sol

OpenAI's customer story with Asana puts hard numbers on GPT-6.1 Sol's production economics. Asana rebuilt its browser agent using OpenAI models through Codex and reports model costs falling by a factor of 76 and runs completing five times faster in testing.

The gains appear to come from a combination of model efficiency and re-architecting the agent with Codex's help, rather than a simple model swap. Asana says the savings are being reinvested to give customers access to more capable models within its product.

The case study is part of a wave of GPT-6.1 Sol deployment news: AWS launched an Ultrafast mode for Sol on Amazon Bedrock targeting latency-sensitive workloads, Bedrock added reasoning summaries for OpenAI models through the Responses API, and Sophos reported a 96% reduction in threat-investigation time using OpenAI's Daybreak. A Korea IT Times analysis argues Sol represents a restructuring of ChatGPT into a work platform rather than a routine model bump.

Caveats: vendor case studies are curated and the 76x figure compares against Asana's own earlier setup, which may have been inefficient. Still, it supports the argument that agent cost — not capability — is the binding constraint, echoing Postman's view that context management is the real production bottleneck.