# GPT-5.6 arrives, locked behind a preview

Sent June 27, 2026

OpenAI previewed GPT-5.6 Sol, Terra, and Luna on June 26, undercutting Anthropic’s flagship by half on price. The catch: it shipped to roughly 20 government-approved partners under a limited preview, so most teams could not use it yet. Claude Mythos 5 still led at 99, with Fable 5 at 95, while Z.AI’s GLM-5.2 entered the top five at 91. The issue examined what the 17-day gap between those launches suggested about the market.

## Leaderboard movers

| Model | Provider | Score | Rank | Movement |
|---|---|---:|---:|---|
| Claude Mythos 5 | Anthropic | 99 overall | #1 | Held #1 |
| Claude Fable 5 | Anthropic | 95 overall | #2 | Held #2 |
| GLM-5.2 | Z.AI | 91 overall | #4 | New to top five |

## Analysis worth opening

- [Fable 5 vs GPT-5.6: Two Bets on Where the Frontier Goes Next](https://benchlm.ai/blog/posts/fable-5-vs-gpt-5-6-market-direction): Anthropic shipped Fable 5 and Mythos 5 broadly at $10/$50. OpenAI previewed GPT-5.6 at half the price, but only a limited group could use it.
- [How We Keep a Benchmark Site Honest](https://benchlm.ai/blog/posts/web-data-pipeline-for-benchmarks): How live benchmark and pricing tables are collected and monitored without maintaining a dedicated scraper for every source.

## This issue in numbers

- Models ranked: 124
- New that week: 3
- Average output price per 1M tokens: $23.92
- Top overall score: 99

## New on BenchLM

- [GPT-5.6 Sol, Terra, and Luna](https://benchlm.ai/models/gpt-5-6-sol): The new family was tracked with pricing and preview notes, pending an independent benchmark suite at general availability.
- [GLM-5.2 entered the top five](https://benchlm.ai/models/glm-5-2): Z.AI’s release reached 91 overall and was the highest-ranked open-weight model on the board.
- [Cost calculator updated](https://benchlm.ai/tools/cost-calculator): The GPT-5.6 tiers were added so readers could price Sol against Mythos 5 using their own token mix.

Archive copy reflects the rankings, prices, and availability stated when this issue was sent. Current pages may show newer evidence.

Canonical page: https://benchlm.ai/newsletter/issues/2026-06-27