Fetching from the wire…
01
02
03
04
05
06
07
08
Top 5 · 2026-06-25 · source-backed
Weeks after launch, Z.ai's open-weight GLM-5.2 now accounts for roughly 75% of all Z.ai model traffic on OpenRouter, with at least one provider serving it past 125 tokens per second (GIGAZINE, citing OpenRouter). The numbers behind the surge: an Artificial Analysis Intelligence Index of 51, the first open model past 80% on Terminal-Bench 2.1, and 62.1 on SWE-bench Pro. Output costs run roughly 5 to 8 times cheaper than Claude Opus 4.8 and about one-sixth of GPT-5.5 Pro. The June 13 suspension of Claude Fable 5 and Mythos 5 poured gasoline on it. People had agentic coding workloads running, the model they were using went away, and GLM-5.2 was sitting right there at a quarter of the price.
Read this next to the Qwen story and the arc is hard to miss. Lab A allegedly harvests agentic-reasoning behavior. Open-weight Lab B ships a model that does agentic coding at 62 on SWE-bench Pro for pennies. I'm not claiming a direct line between those two facts. But the macro pattern is the same: the agentic-coding capability that was a frontier-lab exclusive eighteen months ago is now a commodity you route to by price and latency.
I've started doing this myself in personal projects. Not everything needs Opus. A lot of my agent work is mechanical: write the test, run it, read the failure, patch, repeat. For that loop, a model at 125 TPS and one-sixth the cost changes the math on how aggressively I can fan out. When each subagent is cheap, "spawn 30 of them" stops being a budget decision.
The catch, and I want to be honest here, is that I don't trust leaderboard SWE-bench numbers to predict how a model behaves on my codebase. Benchmark 62 and "good on a 40-file TypeScript repo with weird internal conventions" are different claims. So the move isn't to rip out Opus. It's to set up a real eval on your own traces (see the pass^k story below and the skills section) and let the data decide which workloads drop down to GLM-5.2. Route by evidence, not by leaderboard.
Each link below shares sources, entities, or timing with this story.
Anthropic released Mythos / Shared entities / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (Anthropic released Mythos); both cover Artificial Analysis Intelligence Index, Bench, Claude Opus, GPT; overlapping topics (agentic, coding, model, swe-bench).
output uses Claude Code / Shared entities / Shared topic / What happened next
Linked by a graph relationship (output uses Claude Code); both cover Bench, Benchmark, Claude Opus, Fable; overlapping topics (claude, leaderboard, model, number).
output uses Claude Code / Shared entities / Shared topic / Earlier coverage
Linked by a graph relationship (output uses Claude Code); both cover Bench, Claude Opus, Fable, GLM; overlapping topics (claude, fable, glm-5, model).
output uses Claude Code / Shared entities / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (output uses Claude Code); both cover Bench, Fable, GPT, Mythos; overlapping topics (claude, coding, fable, opus).
Cursor supports Claude Opus / Shared entities / Shared topic / What happened next
Linked by a graph relationship (Cursor supports Claude Opus); both cover Bench, Fable, GPT, Opus; overlapping topics (coding, cost, model, opus).
output uses Claude Code / Shared entities / Shared topic / What happened next
Linked by a graph relationship (output uses Claude Code); both cover Bench, Claude Opus, GLM, GPT; overlapping topics (claude, coding, glm-5, model).
output uses Claude Code / Shared entities / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (output uses Claude Code); both cover Claude Opus, Opus, People, Qwen; overlapping topics (claude, coding, cost, model, number).
Anthropic released Mythos / Shared entities / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (Anthropic released Mythos); both cover Claude Opus, GLM, GPT, Opus; overlapping topics (coding, glm-5, model, open-weight, opus).