OpenAI GPT-5.6-Sol workhorse model beats Fable on agents at $5/$30
GPT-5.6-Sol doubles researcher output tokens, priced 50% less than Fable
Deep Dive
OpenAI released GPT-5.6-Sol, a workhorse model priced at $5/$30 per million tokens ($2.50/$15 for Terra, $1/$6 for Luna). It claims superiority over Fable and Opus on Agents' Last Exam, AA Agent Coding Index, and BrowseComp. Internal use shows average daily output tokens per active researcher more than doubled from the highest level observed for GPT-5.5. Sol excels at agents and coding, while Fable remains smarter but more restricted. Users can get practical tasks done efficiently.
Key Points
- Sol is priced at $5/$30 per million tokens, 50% less than Fable's $10/$50, with Terra and Luna at even lower rates.
- Internal researcher output tokens per day doubled compared to GPT-5.5, and share of compute for coding inference grew 100x in six months.
- Sol outperforms Fable and Opus on agent-focused benchmarks (Agents' Last Exam, AA Agent Coding Index, BrowseComp).
Why It Matters
Sol makes advanced AI cost-effective for practical agents and coding, enabling scaled automation and lower barriers to tool building.