GPT 5.6 is out: is it really better than Claude's Fable 5?
GPT 5.6 is generally available: three sizes (Luna, Terra, Sol), how it scores, and the catch.
Transcript
GPT 5.6 is finally out. But is it actually the best model?
For weeks only early testers had GPT 5.6. This week OpenAI made it generally available to everyone.
It ships in three sizes, Luna, Terra, and Sol. Use it inside the Codex coding tool or the API.
On the independent Artificial Analysis benchmark, Sol scored 59, just behind Fable 5 at 60. That test used Max, not the new Ultra level.
For coding, reviewers praised its long autonomous runs, stronger computer use, and subagents that split work in parallel.
Bottom line: it ran cheaper in one big test. OpenAI calls Sol the strongest, but independent tests give Fable the slight edge.
Quick note: we make Inboxsmith, an AI receptionist that never misses a business call. Get more at inboxsmith dot com.
Sources
Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:
- OpenAI made the GPT 5.6 series of models generally available.(Meet GPT-5.6)
- Before the public release, only early testers had access to GPT 5.6.(So I've been using gpt-5.6 for awhile...)
- GPT 5.6 comes in three sizes: Luna, Terra, and Sol, the largest.(GPT-5.6 Sol vs Fable 5: I Wasn't Expecting This)
- GPT 5.6 can be used in the Codex coding tool, selected with the model command.(GPT-5.6 Sol vs Fable 5: I Wasn't Expecting This)
- In the API, GPT 5.6 is offered as three models: a flagship for agentic work, a balanced model, and a fast, affordable one.(Meet GPT-5.6)
- On the independent Artificial Analysis benchmark, GPT 5.6 Sol scored 59, just behind Fable 5 at 60.(GPT-5.6 Sol vs Fable 5: I Wasn't Expecting This)
- That benchmark tested Sol on the Max setting, not the new Ultra reasoning level.(GPT-5.6 Sol vs Fable 5: I Wasn't Expecting This)
- Ultra is a new reasoning level above Max for GPT 5.6 Sol.(GPT-5.6 Sol vs Fable 5: I Wasn't Expecting This)
- A reviewer reported long autonomous runs of 20-plus hours and much stronger computer use with GPT 5.6.(So I've been using gpt-5.6 for awhile...)
- GPT 5.6 is trained to decide when to delegate to subagents and split work in parallel.(Meet GPT-5.6)
- In one CRM build test, Sol cost less than the same build on Fable 5.(GPT-5.6 Sol vs Fable 5: I Wasn't Expecting This)
- OpenAI's own benchmarks declare Sol the strongest model, while independent tests give the slight edge to Fable 5.(GPT-5.6 Sol vs Fable 5: I Wasn't Expecting This)
- One reviewer estimated spending between 180,000 and 240,000 dollars running GPT 5.6 over about six weeks.(So I've been using gpt-5.6 for awhile...)