GPT-5.6 vs Claude Fable 5
OpenAI's new frontier model against Anthropic's most capable released model — half the price, and a genuine split on the benchmarks both vendors have published.
GPT-5.6 vs Claude Fable 5: side-by-side
| GPT-5.6 (Sol) | Claude Fable 5 | |
|---|---|---|
| Maker | OpenAI | Anthropic |
| Released | Preview 26 Jun 2026 · GA 9 Jul 2026 | First 9 Jun 2026 · global again 1 Jul 2026 (export-control pause) |
| Tier | Flagship | Most capable released model (Mythos-class) |
| Context window | 1,050,000 tokens | 1,000,000 tokens (default and maximum) |
| Max output | 128,000 tokens | 128,000 tokens |
| API price (per 1M) | $5 / $30 | $10 / $50 |
| Cached input | $0.50 (90% off) | Not itemized on Fable 5 page |
| Access | ChatGPT, ChatGPT Work, Codex, API | Claude API, claude.ai, Claude Code, AWS Bedrock, Google Cloud, Microsoft Foundry |
| Safety posture | US government safety review before launch (lifted) | US export-control pull, then reinstatement; bio/chem fallback to Opus 4.8 (<5% of sessions) |
| Standout | Token efficiency — 54% fewer output tokens on agentic coding per OpenAI | State-of-the-art on nearly all tested benchmarks per Anthropic |
Price comparison
| Model | Input (per 1M) | Output (per 1M) |
|---|---|---|
| GPT-5.6 Sol | $5.00 | $30.00 |
| GPT-5.6 Terra | $2.50 | $15.00 |
| Claude Fable 5 | $10.00 | $50.00 |
Estimate your monthly API cost for GPT-5.6 vs Claude Fable 5 with the LLM API pricing calculator →
Published benchmark figures
Figures are each model's own published score as listed on its AI Model Hub page, cross-checked against OpenAI's GA benchmark tally comparing the two.
| Benchmark | GPT-5.6 Sol | Claude Fable 5 | Leader |
|---|---|---|---|
| Terminal-Bench 2.1 | 91.9% | 88.0% | Sol |
| SWE-bench Pro | 64.6% | 80% | Fable 5 |
| Agents' Last Exam | 53.6 | 40.5 | Sol (+13.1) |
| GDPval-AA v2 | Trails (exact score not published) | Leads | Fable 5 |
| Cybersecurity (ExploitGym) | Leads — most capable model yet for long-horizon security tasks | Not the leader on this suite | Sol |
| FrontierMath, HealthBench Professional, AA Intelligence Index v4.1 | Trails | Leads | Fable 5 |
What the table says
- Pricing. Sol is exactly half of Fable 5's rate card on both input and output — the widest clean price ratio between two current frontier-tier models on this site.
- Benchmarks. This is a genuine split, not a rout. Sol's edge is in agentic/terminal tasks and long-horizon cybersecurity work; Fable 5's edge is in software engineering (SWE-bench Pro), knowledge work (GDPval-AA v2) and hard math/science (FrontierMath). Neither vendor's own tally claims a clean sweep.
- Context and output. Nearly identical — 1.05M vs 1M tokens of context, both capped at 128K output tokens.
- Release history. Both models had unusual gating episodes tied to US government action: Fable 5 was pulled under export controls for three weeks, GPT-5.6 spent 12 days under a preview safety review before GA.
Key takeaway
If your workload leans on long-horizon agent tasks, terminal/coding-agent loops or security research, Sol's benchmark lead plus half the price makes it the easy pick. If your workload is software-engineering-heavy (SWE-bench-style multi-file work) or deep knowledge/scientific work, Fable 5's benchmark lead may justify the 2x premium. Run your own token volumes through the pricing calculator to see the monthly difference.
Full GPT-5.6 overview Full Claude Fable 5 overview
Frequently asked questions
Which is cheaper, GPT-5.6 or Claude Fable 5?
GPT-5.6 Sol — $5/$30 per 1M tokens versus $10/$50 for Claude Fable 5: exactly half the price on both input and output.
Which model wins on benchmarks?
It splits. Sol leads Terminal-Bench 2.1, Agents' Last Exam, BrowseComp, OSWorld and cybersecurity. Fable 5 leads SWE-bench Pro, GDPval-AA v2, FrontierMath, HealthBench Professional and the Artificial Analysis Intelligence Index v4.1.
Which has the bigger context window?
GPT-5.6 — 1,050,000 tokens versus 1,000,000 for Claude Fable 5. Both cap output at 128,000 tokens.
Which model is newer?
GPT-5.6 — GA 9 July 2026. Claude Fable 5 first released 9 June 2026, was pulled under export controls, and returned globally 1 July 2026.