Two models landed within a week of each other in June, and they sit at opposite ends of the market. Claude Fable 5 is the most capable model you can call today. GLM-5.2 is the strongest open-weight model you can call today. The interesting question isn't which is "better" in the abstract. It's where each one earns its price.
The ceiling: Claude Fable 5
Anthropic released Fable 5 on June 9 and put it a full tier above Opus 4.8, in what they call the Mythos class. The benchmarks are not close. It leads the Artificial Analysis Intelligence Index at 60, and scores 80.3% on SWE-Bench Pro, with the next-best model eleven points behind, and its agentic score of 80.7 is the highest anyone has posted. Across the boards that track these things, Fable 5 ranks first for overall intelligence, first for coding, first for agentic work. It has a million-token context and can return up to 128K tokens in one response. This is the model you reach for when an agent has to run for hours, or a migration spans dozens of files and one wrong turn is expensive.
The catch is the bill. Fable 5 runs $10 per million input tokens and $50 per million output, which is double Opus 4.8. One write-up put it bluntly: roughly twice the cost for about six percent more performance over the previous top model. If you need that six percent, it's worth every cent. Most workloads don't.
The value play: GLM-5.2
GLM-5.2 came out on June 13 and topped the Artificial Analysis Intelligence Index among open-weight models at 51 — fourth overall, behind only Fable 5, Opus 4.8 and GPT-5.5 — clear of DeepSeek V4-Pro, MiniMax-M3 and Kimi K2.6. On coding it took first on Design Arena, scored 62.1 on SWE-bench Pro (against DeepSeek V4-Pro's 55.4), and beat GPT-5.5 outright. It's a 744B-parameter MoE with about 40B active, a million tokens of context, and Z.ai is releasing the weights under MIT. (We go deeper in the GLM-5.2 deep dive.)
It is not going to out-score Fable 5 on the hardest agentic benchmarks. It doesn't need to. It gets you frontier-adjacent coding at a small fraction of Fable 5's price, and you can run it yourself if you want to.
How to choose
Use Fable 5 when the task is genuinely at the frontier: multi-day autonomous sessions, large risky migrations, the cases where the top few percent of capability changes the outcome and the token bill is a rounding error next to an engineer's time.
Use GLM-5.2 for everything else, which is most things: day-to-day code generation, refactors, agent loops, anything you run at volume. You keep the bulk of the quality and cut the cost by a large multiple.
The good news is you don't have to commit. On Turiloop both sit behind one OpenAI-compatible key, so you can default to GLM-5.2 and escalate the rare hard job to Fable 5 by changing the model field:
def solve(prompt, hard=False):
model = "claude-fable-5" if hard else "glm-5.2"
r = client.chat.completions.create(
model=model,
messages=[{"role": "user", "content": prompt}],
)
return r.choices[0].message.contentThat pattern, cheap by default and frontier on demand, is usually the right answer. Pay-as-you-go, international card, no Chinese phone number. Get a key on Turiloop.