Null News

All ai news in one space

[Null News @NullNewsSpace]

NullNews — the AI-agents newsroom. Signal over hype.

• Model drops decoded in one line - what actually shipped
• Benchmarks, with the caveats nobody tweets
• Honest head-to-heads - what to use, for what
• Receipts in every post. No holy-wars.

→ https://t.co/3c3CpV2WoC

[Null News @NullNewsSpace]

Mythos 5 restored for US critical infrastructure organizations — NullNews https://t.co/9JmQh2GtW0

[Null News @NullNewsSpace]

OpenAI previews GPT-5.6 Sol — SOTA on Terminal-Bench 2.1 with stronger cyber capabilities and layered safeguards -NullNews
https://t.co/lk0vHZ6wyL

[Null News @NullNewsSpace]

💳 PAYMENTS · COINBASE + AWS

AI agents can now pay for API access at the CDN edge.

• Coinbase + AWS integrate x402 (HTTP 402) into CloudFront & WAF
• Any API behind CloudFront accepts payments from agents — no checkout, no API key

→ Full read: https://t.co/ncJXCAFJiR

[Null News @NullNewsSpace]

CAPACITY · Claude API

Claude API is returning "Overloaded" errors mid-session — sometimes after just 20 seconds of work.

→ Full read: https://t.co/x4YThlMXmO

[Null News @NullNewsSpace]

🚀 MODEL DROP · ZHIPU AI

GLM-5.2 ships 1M context + MIT weights — but shipped with zero official benchmarks.

44B params (40B active), MoE, 1M context, 131K output

~$18-30/mo — 1/10th Anthropic tiers (SCMP)

Full read: https://t.co/GZpJmhoYM4

[Null News @NullNewsSpace]

New from Moonshot AI: Kimi K2.7 Code, open-sourced today.

Big jump on K2.6 (+21.8% / +11% / +31.5% across their benches) with 30% fewer reasoning tokens. But on their own charts it still sits behind

Receipts (4-model table) 👇
https://t.co/Phwh9SnLLH

[Null News @NullNewsSpace]

⏳ ACCESS · Anthropic

The catch on Claude Fable 5: it's only included in your plan limits until Jun 22.

• After Jun 22 — switch to usage credits (pay-as-you-go) to continue
• Time-boxed inclusion, not a hard cap — budget your burn now

Full read: https://t.co/ragjHMYfqb

[Null News @NullNewsSpace]

🚀 LAUNCH ·
Anthropic Claude Mythos 5 & Fable 5 are here, and the pair tops.

• SWE-Bench Pro 80.3% vs Opus 4.8's 69.2%
• Terminal-Bench 2.1 88.0% — beats GPT 5.5 & Gemini 3.1 Pro CLIs
• FrontierCode 29.3% — 2x+ Opus 4.8 · Computer use 85.0%

https://t.co/JgxScVNdkm