Anthropic Releases Claude Opus 4.8 and Fable 5; Z.ai Unveils GLM-5.2 Among Next-Gen Models

모델 출시 | Wed Jun 17 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 5 sources

Anthropic, Z.ai, and JetBrains released new models specialized for coding, agents, and long-horizon tasks.

Analysis

[Anthropic] released Claude Opus 4.8 ^[1]

Improvements across benchmarks compared to Opus 4.7
Introduced effort control feature and Claude Code dynamic workflows
Fast mode at 2.5x speed with 3x price reduction
Surpassed GPT-5.5 with 84% on Online-Mind2Web

[Anthropic] unveiled Claude Fable 5 and Mythos 5 ^[2]

Mythos-class top-performing model
SOTA in software engineering and scientific research
Priced at $10/$50 per 1M input/output tokens
Mythos 5 provided to cyber defenders via Project Glasswing

[Z.ai] announced GLM-5.2 open weights model ^[3]

Ranked #1 in open weights with 51 on Artificial Analysis Intelligence Index
744B total parameters with 40B active
1M token context
Tied with GPT-5.5 at 1524 points on GDPval-AA v2
MIT license

[Z.ai] revealed GLM-5.2 architecture optimized for long-horizon coding ^[4]

IndexShare reduces per-token FLOPs by 2.9x at 1M context
Improved MTP layer boosts speculative decoding acceptance length by 20%
81.0 on Terminal-Bench 2.1
62.1 on SWE-bench Pro
Within 1% of Opus 4.8 on FrontierSWE

[JetBrains] released Mellum2 12B MoE open source model ^[5]

12B total parameters with 2.5B active per token in MoE structure
Inference more than 2x faster compared to similarly sized models
Optimized for routing
RAG
sub-agents
and code completion
Apache 2.0 license

Sources