Published: 2026-07-18
Analysis & perspective

GPT-5.6 Soul vs Fable 5 vs Grok 4.5 vs GLM 5.2: How They Compare

Name: GPT-5.6 Soul vs Fable 5 vs Grok 4.5 vs GLM 5.2: How They Compare
Uploaded: 2026-07-18
Description: A plain-language rundown of four flagship models from mid-2026 and what each is tuned for.

Chapters / key moments (click to jump — plays here on the page)

Four flagship models landed within weeks of each other in mid-2026 — OpenAI's GPT-5.6 Soul, Anthropic's Fable 5, xAI's Grok 4.5, and Z.ai's open-weight GLM 5.2. Julian Goldie's rundown covers what each is tuned for, where to access it, and the trade-offs. The honest verdict: none wins outright, they're closer than the headlines suggest, so pick by task — and always check the output yourself rather than trusting a model just because it looks finished.

Source video

"GPT-5.6 Sol VS Fable 5 VS Grok 4.5 VS GLM 5.2: Who Wins?" by Julian Goldie SEO — Watch on YouTube →

The Four Models at a Glance

GPT-5.6 Soul (OpenAI, live July 9, 2026). Top of a three-tier family — Soul, then Terra, then the light/fast Luna. Strong at coding and "design judgment": hand it a rough brief and it builds a clean layout, then critiques and tidies its own work. An Ultra setting runs four helpers in parallel on one big task. Access: ChatGPT, Codex, and the OpenAI API.
Fable 5 (Anthropic). Launched June 9, had access suspended June 12 to follow US export rules, and returned July 1, 2026. Its lead grows the longer and harder the task; strong at vision (rebuild code from a screenshot). Anthropic says a small slice of requests — mostly cyber-security and biology, under ~5% of sessions — is handed to the smaller Opus 4.8 model.
Grok 4.5 (xAI, July 8, 2026). Trained alongside the Cursor coding tool. The headline is efficiency — fast, using roughly half the tokens of some rivals for the same task. It's the default in Grok Build and available in Cursor on every plan. EU availability lagged to around mid-July, so check your region.
GLM 5.2 (Z.ai, June 13, 2026; open weights June 16). The one you can own — an MIT-licensed open-weight mixture-of-experts model with a 1M-token context you can download, run yourself, and fine-tune. Billed as the strongest open coding model: 81 on Terminal-Bench 2.1 and 62.1 on SWE-bench Pro (which the video says beats GPT-5.5 on that real-world bug-fixing test). Plugs into tools like Claude Code and Cline with a quick settings swap. Caveat: the hosted cloud version routes data through servers in China, so many teams self-host.

Key Takeaways

Soul — the all-rounder with the best eye for design and clean coding.
Fable 5 — reach for it when the task is long and complex and needs to stay focused for hours.
Grok 4.5 — the speed pick: quick, token-efficient, great for building an app fast.
GLM 5.2 — the one you own: open weights, huge context, self-hostable with nobody else in the loop.
Biggest structural split is openness. Three are closed models you reach through their makers; GLM 5.2 is the one you can actually download and keep.
Bottom line: on everyday work all four give strong results — none is bad. Pick by task, and verify the output yourself.

GPT-5.6 Soul vs Fable 5 vs Grok 4.5 vs GLM 5.2: How They Compare

The Four Models at a Glance

Key Takeaways

More ChatGPT news

Go deeper: ChatGPT guides