GPT-5.6 Sol Pressures Anthropic With Near-Fable 5 Benchmarks at Lower Cost

OpenAI’s GPT-5.6 Sol scores 59 points on Artificial Analysis’s Intelligence Index, just behind Claude Fable 5 at 60. The result positions Sol close to Anthropic’s top model on aggregated benchmarks while giving OpenAI a strong performance story for its new flagship model.

In the Coding Agent Index, Sol reaches 80 points when running in OpenAI’s Codex environment, ahead of every other model listed in the article. In AA-Briefcase office-task testing, Sol earns the highest “Presentation Elo,” though Fable 5 still leads the overall ranking.

Sol costs $1.04 per task, compared with $2.75 for Claude Fable 5, according to the cost-per-task chart cited in the article. That puts Sol at about one-third of Fable 5’s task cost while staying close on the Intelligence Index.

OpenAI’s smaller GPT-5.6 variants are cheaper still: Terra is listed at $0.55 per task and Luna at $0.21 per task. For teams comparing frontier-model performance against budget pressure, the key takeaway is that benchmark gaps may matter less when cost gaps are this large.

GPT-5.6 introduces a cache-write fee and discounts cache reads by 90 percent, according to the article. Token prices per million are listed as $5 input and $30 output for Sol, $2.50 input and $15 output for Terra, and $1 input and $6 output for Luna.

The article also says Sol uses fewer output tokens than similarly performing models, with OpenAI CEO Sam Altman cited for a figure of up to 54 percent fewer output tokens in agentic coding tasks. If that holds across real workloads, customers would evaluate not just sticker prices but total task costs.

The broader implication is competitive pressure on Anthropic. The article frames OpenAI as still pricier than some competitors, including Chinese open models, Meta’s Muse 1.1 and xAI’s Grok 4.5, but now attacking Anthropic with a combination of benchmark strength and lower per-task costs.

For buyers, the practical move is to compare models by task-level cost, coding performance, output-token usage and workflow fit rather than leaderboard rank alone. For AI labs, the risk is that faster price cuts could squeeze margins across the industry.