Comparison
GLM-5.2 vs Opus 4.8: which should you actually use?
With Fable 5 suspended, the real decision for most builders isn't exotic — it's this: the open value pick (GLM-5.2, which you can run yourself) or the stay-in-Claude option (Opus 4.8, which was never gated). Here is the honest, sourced read, with a pick by job.
Quick summary
- GLM-5.2 is open-weight (MIT) and costs roughly a sixth of Opus 4.8 on output — and because the weights are public, no directive can switch your copy off.
- Opus 4.8 edges it on coding benchmarks and is the smoothest path if your work already lives in Claude or Claude Code.
- For many people the answer is both: GLM-5.2 as the everyday floor, Opus 4.8 as the stay-in-Claude option to fail over to.
The numbers, side by side
Prices are per million tokens; benchmark is Terminal-Bench 2.1 (agentic coding). Sources below.
| GLM-5.2 | Opus 4.8 | |
|---|---|---|
| Price · input | $1.40 | $5.00 |
| Price · output | $4.40 | $25.00 |
| License | Open weights (MIT) | Closed · API only |
| Run it yourself? | Yes | No |
| Context window | 1M tokens | Up to 1M (max) |
| Coding · Terminal-Bench 2.1 | 81.0 | 85.0 |
| Available today | Yes | Yes |
Where GLM-5.2 wins
Cost, by a wide margin. At about $1.40 in and $4.40 out per million tokens against Opus 4.8's $5 and $25, GLM-5.2 is roughly three-and-a-half times cheaper on input and close to six times cheaper on output. At any real volume, that is the difference between a side project and a line item.
And ownership. GLM-5.2 ships open-weight under an MIT license, so you can run it on your own hardware or a cheap host — which, after the month the frontier just had, matters: a model whose weights you hold can't be switched off by a directive. On quality it is not a downgrade you'd flinch at either — it lands a few points behind Opus on coding benchmarks and ahead of GPT-5.5 on SWE-bench Pro.
The honest caveat: using GLM through Z.ai's hosted API raises data-handling questions for some teams. Because the weights are open, self-hosting is the way around that — at the cost of running the infrastructure yourself.
Where Opus 4.8 wins
Top-end quality and Claude continuity. Opus 4.8 edges GLM-5.2 on the coding benchmarks, and for the hardest agentic and reasoning tasks that margin can be worth the price. More importantly, if your work already lives in Claude Code with prompts and review steps tuned for Claude, Opus is a drop-in — no re-tooling, no new infrastructure, and enterprise data terms out of the box.
It is also simply available. Opus 4.8 was never part of the June suspension, so unlike Fable 5 it is something you can use in full today, through the same Claude surfaces you already know.
Pick by your job
- Cost-sensitive or high-volume API workGLM-5.2 — the price gap compounds fast.
- You want to own it / avoid revocation riskGLM-5.2, self-hosted on the open weights.
- Your work lives in Claude Code or a Claude-tuned flowOpus 4.8 — drop-in continuity, no re-tooling.
- The hardest coding or agentic tasks, top quality firstOpus 4.8 — but test GLM-5.2 on your own task before paying the premium.
- You want resilience against the next directiveBoth — GLM-5.2 as the floor, Opus 4.8 as the stay-in-Claude failover.
The honest recommendation
1
Default to GLM-5.2. For most everyday coding and automation it is close enough to Opus at a fraction of the cost — and you can hold the weights. Start here unless you have a reason not to.
2
Keep Opus 4.8 for the hard edge. The toughest tasks and Claude-native workflows are where its lead earns the price — and it doubles as the stay-in-Claude option you route to.
3
Decide on your task, not a leaderboard. Run both on one real job of yours and compare. Not sure how to wire it up? Ask the advisor for a setup matched to your work and budget.
Need the short answer?
Fable 5 is back worldwide as of July 1 — but capped at 50% of your weekly limit until July 7. See the live status, or use GLM-5.2 or the new Sonnet 5 for cheaper work.
Read the brief Fabel 5 spelling guideFAQ
Is GLM-5.2 as good as Opus 4.8?
Close on coding: it scores 81 to Opus's 85 on Terminal-Bench 2.1, and edges GPT-5.5 on SWE-bench Pro. Opus is still ahead at the top end, but GLM-5.2 is within a few points at roughly a sixth of the output cost.
Can I run GLM-5.2 myself?
Yes. It is open-weight under an MIT license, so you can self-host on your own GPUs or a cheap host — which also means no directive can take your copy offline. Opus 4.8 is closed and API-only.
Which is cheaper?
GLM-5.2, by a lot: about $1.40 / $4.40 per million input / output tokens versus Opus 4.8 at $5 / $25 — roughly a sixth of the cost on output.
Track what's worth using now
Fable 5 is back — get one email when the next big move lands (Fable pricing, Sonnet 5, GPT-5.6), plus the occasional practical update. No spam, leave anytime.
Sources
This page is independent. Official provider pages are the source of record for access, pricing, and policy.