Fresh head-to-head reviews pit OpenAI's GPT-5.6 Sol, holder of an 88.8% score on a leading coding benchmark, against Anthropic's Claude Fable 5 and its 80.3% software engineering mark.
Key Points:
- GPT-5.6 Sol tops Terminal-Bench 2.1 at 88.8%, and its Ultra mode pushes the score to 91.9%.
- Claude Fable 5 keeps the widest published lead on SWE-Bench Pro at 80.3%, versus 58.6% for GPT-5.5.
- Sol remains in a limited government-approved preview, while Fable 5 returned to global availability on Jul. 1.
GPT-5.6 Sol Benchmark Claims
OpenAI previewed the GPT-5.6 family on Jun. 26, its first release since GPT-5.5 in April, splitting the line into three tiers with Sol as the flagship.
The company says Sol reaches 88.8% on Terminal-Bench 2.1, a test of command-line coding agents that plan, iterate and coordinate tools. A compute-heavy Ultra mode, which spins up coordinated subagents to accelerate complex work, stretches that figure to 91.9%, the top published mark on the Terminal-Bench chart.
Reviewers who compared the published charts place Fable 5 several points behind Sol on the same terminal test, though cited figures vary between 83.4% and 84.3%. On the ExploitBench security suite, Sol reportedly matches Mythos-class performance while spending roughly one third of the output tokens, a cost compression that matters in long agent runs.
Almost nobody outside the preview can verify those numbers independently yet, a caveat several reviewers flagged while acknowledging the raw scores.
Also Read: OpenAI And Anthropic Want SpaceX-Sized IPOs, But Wall Street May Choke
Fable 5 Coding Lead And Pricing
Fable 5 still owns the benchmark most reviewers treat as decisive for autonomous software work, and its edge there is not small. It scores 80.3% on SWE-Bench Pro, which measures end-to-end fixes of real GitHub issues, against 58.6% for the older GPT-5.5, and OpenAI has published no GPT-5.6 figure there.
Analysts who found gaps of that size across coding, reasoning and knowledge tests doubt a single incremental release can close them fully.
Price cuts the other way, since Sol is reportedly listed at $5 per million input tokens and $30 for output, half of Fable 5's $10 and $50. Several reviewers argued that the sensible setup routes terminal-driven agents toward Sol, once it opens up, and repository-level fixes toward Fable 5.
Access draws the sharpest line, since Sol remains in a limited preview for roughly 20 government-cleared partners, while Fable 5 returned worldwide on Jul. 1 with a temporary usage bonus for paid subscribers through Jul. 7.
June turned frontier model access into a moving target for both laboratories, and that whiplash frames every review. Washington forced Fable 5 and its more powerful sibling Mythos 5 offline on Jun. 12, citing severe cybersecurity risks, after Amazon researchers surfaced a jailbreak that produced exploit code. Commerce Secretary Howard Lutnick confirmed the reversal on Jun. 30 following a two-week review, days after Mythos 5 quietly returned to about 100 vetted American organizations.
Read Next: Why Is ETH Still Weak While Ethereum Staking Hits Record Highs?