Amplitude's Wave agent uses OpenAI models to ship product improvements in 1–2 days

• Amplitude's proactive product agent, Wave, identifies opportunities, generates implementation plans, and coordinates with coding agents to ship improvements in 1–2 days, down from roughly a month.
• Wave is approximately 80% powered by OpenAI models, including GPT‑6.1 Sol, GPT‑6 Luna, and GPT‑Image‑2.5 Flare, which are assigned to specific tasks based on reasoning needs, latency, and cost.
• In a test on Amplitude's homepage, Wave personalized suggestion cards based on user context, resulting in a 160% lift in card-driven sends relative to control after one day.
• GPT‑6.1 Sol scored 4.06 out of 5 in opportunity investigation evaluations and had a mean cost of $1.28 per dispatch in data science agent tests, compared to $2.63 for a competing model.

Model Performance and Cost

• GPT‑6.1 Sol led all five investigation criteria with a median investigation time of 54 minutes.
• GPT‑6 Luna handles high-volume, lower-cost tasks like code scanning to keep latency and inference costs low.
• GPT‑Image‑2.5 Flare scored 80.3 out of 100 for solution wireframes, outperforming another model's 63.9 score.

Workflow and Automation

• Teams control automation levels, sending bug fixes directly to coding agents while keeping human review for new features.
• Among customer-launched experiments, one in four went live the same day Wave surfaced the opportunity, and two in three within a week.