Our July verdict: GPT-5.6 Sol was the best coding model, and Fable 5 kept the edge on writing. October update: both have been passed. Claude Opus 5.5 is now our pick for the best AI model, and GPT-6 Sol replaces GPT-5.6 Sol as OpenAI's everyday default at $2/$10.
OpenAI released GPT-5.6 Sol on July 9, 2026, and it took the top coding scores from Claude Fable 5 at about half the price. Sol Ultra hit 91.9% on Terminal-Bench 2.1, and Sol led the Artificial Analysis Coding Agent Index by 2.8 points.
The July 9 scoreboard
Terminal-Bench 2.1 runs agentic coding tasks in a real terminal, and the Coding Agent Index rolls several coding evals into one score. Sol took both, and at launch it cost $5/$30 per million tokens against Fable 5's $10/$50.
Where things stand in October
- #1 on the Artificial Analysis Intelligence Index
- Leads SWE-bench Pro at 89.9%
- Hard coding and research
The catch: twice the price of Sonnet 5.5.
- Released September 22
- Half the price of GPT-5.6 Sol
- Replaces both Sol and Terra
The catch: OpenAI's top model is GPT-6 Astra.
- Released September 28
- Built for production workloads
- Same price as GPT-6 Sol
The catch: step up to Opus for the hardest jobs.
- Replaced Fable 5, same price
- Brand voice and long-form copy
- Keeps Fable 5's prose edge
The catch: 2.5 times the price of Opus 5.5.
- Released September 22
- Half the price of GPT-5.6 Luna
- High-volume routine tasks
The catch: not meant for frontier-level work.
- Led coding benchmarks on July 9
- About a million tokens of context
- Sol Ultra variant for peak scores
The catch: GPT-6 Sol does the job for half.
Frontier models at a glance
Prices checked October 15, 2026, on each company's own pricing page.
| Model | Price per 1M tokens (in / out) | Status | Best for |
|---|---|---|---|
| Claude Opus 5.5 | $4 / $20 | #1 overall since Sep 22 | Hard coding and research |
| Claude Sonnet 5.5 | $2 / $10 | Current, Sep 28 | Everyday production work |
| Claude Fable 5.1 | $10 / $50 | Current, replaced Fable 5 | Brand voice, long-form writing |
| GPT-6 Astra | $10 / $50 ($20 / $75 past 272K) | Current, Sep 3 | Science and computer automation |
| GPT-6 Sol | $2 / $10 | Current, Sep 22 | OpenAI's everyday default |
| GPT-6 Luna | $0.10 / $0.50 | Current, Sep 22 | High-volume, cheap tasks |
| GPT-5.6 Sol | $4 / $20 | Previous generation | Switch to GPT-6 Sol |
| GPT-5.6 Terra | $2 / $12 | Previous generation | Switch to GPT-6 Sol |
| GPT-5.6 Luna | $0.20 / $1.20 | Previous generation | Switch to GPT-6 Luna |
| Gemini 4 Argon | $2 / $10 through Nov 30, then $4 / $20 | Current | Google's top model |
| Grok 4.7 | $2 / $6 (doubles past 200K) | Current | xAI's flagship |
Who won what in July
| Category | GPT-5.6 Sol | Claude Fable 5 |
|---|---|---|
| Top Terminal-Bench 2.1 score | ✓ | ✗ |
| Coding Agent Index lead | ✓ | ✗ |
| Lower launch price | ✓ | ✗ |
| About 1M tokens of context | ✓ | ✓ |
| Writing and brand voice | ✗ | ✓ |
| Hardest long-document reasoning | ✗ | ✓ |
| Consistent over many edit rounds | ✗ | ✓ |
Scores come from the OpenAI and Anthropic launch posts; the last three rows are our July ranking.
GPT-5.6 Sol: July's coding leader
GPT-5.6 comes in three tiers: Sol is the flagship, Terra the balanced model and Luna the cheapest. Sol launched at $5/$30 per million tokens, and OpenAI cut it to $4/$20 for three months starting August 21.
Sol Ultra is the setting that holds the 91.9% Terminal-Bench score. It runs four agents in parallel by default, trading more tokens for better results. All three tiers have about a million tokens of context, with higher rates past 272K.
Claude Fable 5: still the better writer
Sol's writing is competent but clinical. Fable 5 produced the most natural prose of any frontier model in mid-2026, held up better on the hardest long documents and drifted less across rounds of edits.
Anthropic reports 88.0% on Terminal-Bench 2.1 for Mythos 5 and Fable 5, the higher of the two models' scores. OpenAI's own table puts Fable 5 at 83.1%. Fable 5.1 has since replaced it at the same $10/$50.
Terra and Luna: the value tiers
Terra launched at $2.50/$15 and Luna at $1/$6. On July 30, OpenAI cut Terra by 20% to $2/$12 and Luna by 80% to $0.20/$1.20.
GPT-6 now beats both on price: GPT-6 Sol costs $2/$10 and GPT-6 Luna $0.10/$0.50. See our cheapest AI model verdict.
What replaced them
In September, OpenAI shipped GPT-6 Astra on September 3 and the cheaper GPT-6 Sol and Luna on September 22. Anthropic shipped Claude Opus 5.5 on September 22 and Sonnet 5.5 on September 28.
Opus 5.5 is now #1 on the Artificial Analysis Intelligence Index and leads SWE-bench Pro at 89.9%. Our September frontier ranking has the full comparison.
What about Gemini and Grok?
Gemini 4 Argon is Google's top model at $2/$10 through November 30, then $4/$20, and it's the natural pick if you live in Workspace. Grok 4.7 is xAI's flagship at $2/$6, with rates doubling past 200K tokens.
Which model should you use?
The verdict
GPT-5.6 Sol won July on coding and price, and Fable 5 kept the writing crown. In October, use Claude Opus 5.5 for hard work, GPT-6 Sol or Sonnet 5.5 for everyday jobs, Fable 5.1 for writing, and an open-weight model you run yourself (see our best free local LLM verdict) for private data.
FAQ
What is GPT-5.6 Sol?
GPT-5.6 Sol was OpenAI's flagship model when it launched on July 9, 2026, the top tier of the GPT-5.6 family alongside Terra and Luna. It now costs $4 per million input tokens and $20 per million output. It has since been replaced at the top by GPT-6 Astra (September 3) and the cheaper GPT-6 Sol (September 22, $2/$10), and Claude Opus 5.5 now leads the overall ranking.
Is GPT-5.6 Sol better than Claude Fable 5?
On coding benchmarks in July, yes. Sol Ultra scored 91.9% on Terminal-Bench 2.1 vs 88.0% for Claude Mythos 5 and Fable 5, and Sol led the Artificial Analysis Coding Agent Index by 2.8 points. Fable 5 kept an edge on writing quality. Both have since been overtaken: Claude Opus 5.5 is now #1 on the Artificial Analysis Intelligence Index and leads SWE-bench Pro at 89.9%.
How much does GPT-5.6 cost?
Per million tokens, OpenAI now lists GPT-5.6 Sol at $4 input / $20 output, Terra at $2 / $12 and Luna at $0.20 / $1.20, with higher rates once a prompt passes 272K tokens. The newer GPT-6 Sol costs $2 / $10 and GPT-6 Luna $0.10 / $0.50, so most teams still on GPT-5.6 should switch to GPT-6.
Should I switch from Fable 5 to GPT-5.6 Sol?
Not anymore. Both have been superseded. For agentic coding, test Claude Opus 5.5 ($4/$20) first. For everyday production work, use Claude Sonnet 5.5 or GPT-6 Sol, both $2/$10. Keep Fable 5.1 only for writing-heavy, brand-voice work. Route by workload, not by brand.
What are Terra and Luna?
Terra and Luna are the mid and efficient tiers of GPT-5.6. OpenAI now lists Terra at $2/$12 and Luna at $0.20/$1.20 per million tokens. Both are now outclassed on price by GPT-6: GPT-6 Sol costs $2/$10 and GPT-6 Luna costs $0.10/$0.50, so new projects should start on GPT-6.