ImportantModels ·
OpenAI Astra Hits Critical Cybersecurity Tier; Dev Paused
OpenAI's Astra can no longer rule out critical cyber capabilities, a frontier-lab first. Development paused, five safeguards enacted.
Read the full callNews section
Models
Foundation and frontier model releases, updates, and capability changes.
Edition of Saturday, August 8, 2026All news
18 Important · 3 verdict changes · 9 FYI · 0 watching · 30 total
ImportantModels ·
Microsoft Ships MAI-Cyber-1-Flash, Hits 96% on CyberGym
MAI-Cyber-1-Flash scores 96% on CyberGym (12 points above Mythos 5 at half the cost) alongside Project Perception, an agentic security platform launching Aug 3.
ImportantModels ·
Kimi K3 Weights Expected Today: 2.8T Free Download
Moonshot AI is expected to release the 2.8T-parameter Kimi K3 weights today under a Modified MIT license. A Hugging Face countdown was still ticking. The largest open-weight release in history is hours away.
Q
+3ImportantModels ·
Qwen3.8-Max Ships: Alibaba Claims Fable 5 Parity at $2/$6
Qwen3.8-Max claims Fable 5 parity on coding and reasoning at $2/$6 per 1M tokens, 5-8x cheaper. Open weights ship Aug 10. The fourth Chinese frontier model in six weeks.
ImportantModels ·
DeepSeek V4 Ships: Liang Wenfeng's Reported AGI-First Bet
DeepSeek V4 launched mid-July with two MIT-licensed variants at commodity pricing. A circulated investor transcript reportedly reveals founder Liang Wenfeng's AGI-first strategy: open source forever, profit later.
ImportantModels ·
GPT-5.6 Luna Drops 80%, Terra 20%: Sol Self-Optimized
OpenAI cut GPT-5.6 Luna 80% and Terra 20%. Sol autonomously rewrote production GPU kernels, the first frontier model to self-optimize its serving costs and pass the savings to customers.
ImportantModels ·
Claude Opus 5 Ships at $5/$25: Near-Fable 5 Intelligence
Anthropic shipped Claude Opus 5 on July 24 at $5/$25 per 1M tokens, near-Fable 5 intelligence at half the price, with SOTA Frontier-Bench scores and 85% fewer safety interruptions. Our verdict: Recommended.
ImportantModels ·
Three Gemini Flash Models Ship: 3.5 Pro Still Delayed
Google shipped three efficiency-focused Gemini Flash models, while the flagship 3.5 Pro misses its June target and Gemini 4 pretraining begins.
D
ImportantModels ·
Laguna S 2.1: The West's Open-Weight Answer Arrives
Poolside's 118B MoE open-weight model beats DeepSeek-V4-Pro-Max and Inkling on Terminal-Bench 2.1 at 8B active params. $0.10/1M tokens, runs on a single DGX Spark.
ImportantModels ·
Fable 5 Gets Subscription Access: Max/Team Premium Included at 50%
Anthropic reversed course: Max/Team Premium plans now include Fable 5 at 50% limits starting July 20, the fourth access-policy change in four weeks. Pro and Team Standard get a one-time $100 credit.
+1ImportantModels ·
Kimi K3 Tops Arena Coding at $3/$15, Open Weights July 27
Moonshot AI's Kimi K3 is a 2.8T open-weight model that beat Fable 5 on Arena Code WebDev at $3/$15, the third Chinese frontier release this month.
ImportantModels ·
Thinking Machines Ships Inkling, 975B Open-Weights MoE
Thinking Machines Lab's first model Inkling ships as a 975B Apache 2.0 MoE with controllable thinking effort: built to be customized, not to win benchmarks.
ImportantModels ·
Fable 5 Extended to July 19: Second Extension in 5 Days
Anthropic pushed the Fable 5 credit-cliff to July 19, the third deadline and second extension in five days. GPT-5.6 Sol at half the price is the pressure behind it.
+1ImportantModels ·
Meta Enters AI Coding Race With $1.25/M Muse Spark 1.1
Meta shipped Muse Spark 1.1 on July 9, 2026: its first paid AI model at $1.25/$4.25 per million tokens, undercutting GPT-5.6 Sol's output by ~86%.
ImportantModels ·
GPT-Live Ships Full-Duplex Voice on ChatGPT
OpenAI launched GPT-Live-1 and GPT-Live-1 mini, full-duplex voice models that listen and speak simultaneously. The third generation of ChatGPT voice is here.
ImportantModels ·
Grok 4.5 Ships at $2/$6: Undercuts Opus by 60%
SpaceXAI released Grok 4.5 at $2/$6 per million tokens, undercutting Opus 4.8 by 60% and GPT-5.6 Sol by 80%. The most aggressive frontier-adjacent pricing this year.
ImportantModels ·
Anthropic Extends Fable 5 to July 12: Cliff Postponed
Anthropic extended Fable 5 subscription access through July 12, hours before the midnight credit-only cutoff. The cliff moved five days, with GPT-5.6 Sol launching Thursday at half the price.
ImportantModels ·
GPT-5.6 Goes Public July 9: Government Gate Opens
GPT-5.6 Sol launches publicly July 9 after Commerce Department clearance. The 12-day government gate is open, and our verdict moves to Recommended.
Verdict changedModels ·
GPT-5.6 Terra & Luna Hit GA: Verdict Moves to Conditional
OpenAI's GPT-5.6 Terra ($2.50/$15) and Luna ($1/$6) exited government-gated preview on July 9, 2026, and are now generally available via API and Codex. Both move from pending to conditional. Pricing held stable through the GA transition, but all benchmarks remain vendor-reported and standard ChatGPT access is still absent.
Verdict changedModels ·
GPT-5.6 Sol Beats Fable 5 on DeepSWE at One-Third the Cost
GPT-5.6 Sol scored ~72% on the DeepSWE coding-agent benchmark to Claude Fable 5's ~70%, at one-third the cost. The cheaper model leads the first credible third-party comparison since both reached GA.
Verdict changedModels ·
GPT-4o Goes Conditional: Delisted from OpenAI Pricing
GPT-4o was removed from OpenAI's API pricing page, superseded by the GPT-5.x family. Our verdict moves from Recommended to Conditional. It's legacy, not current-gen.
FYIModels ·
Fable 5 Biology Fallbacks Drop 85% After Classifier Rewrite
Fable 5's biology safety classifier was rewritten: false-positive fallbacks dropped ~85%, directly addressing the #1 user complaint since the July 1 restoration.
GFYIModels ·
Grok 4.6 Set for Aug 7, Grok 4.7 Weeks Later
Elon Musk confirmed Grok 4.6 launches around August 7, with Grok 4.7 following weeks later. Two frontier models in one month is a cadence no other AI lab has matched.
FYIModels ·
GPT-Transcribe Replaces Whisper: 52% Fewer Errors, 25% Less
OpenAI shipped GPT-Transcribe and GPT-Live-Transcribe to the API, replacing Whisper with 52% fewer errors and a 25% price cut to $4.50 per 1,000 minutes. Both accept context prompts, keyword hints, and language hints for dramatically better accuracy on specialized audio.
FYIModels ·
Kimi K3 Open Weights Now Live on Hugging Face
Moonshot AI released the full Kimi K3 open weights (2.8T parameters) on Hugging Face on July 27 under the Kimi K3 License, a custom license with commercial thresholds. Self-hosting eliminates China data-sovereignty risk for regulated teams.
C+2FYIModels ·
Claude Opus 5 Leak Shows Fable 5-Class 3D Rendering
Claude Opus 5 has surfaced across multiple providers with 3D rendering demos that exceed Fable 5 in quality. Anthropic has not confirmed anything, and pricing remains unannounced.
FYIModels ·
Fable 5 Disproves Jacobian Conjecture, Researcher Says
Fable 5 disproved the 87-year-old Jacobian Conjecture. Anthropic researcher Levent Alpöge found a counterexample, verified in Lean within hours.
FYIModels ·
Gemini 3.5 Pro: Fourth Delay, No GA Date
Gemini 3.5 Pro is "months behind schedule", its fourth delay since May. Coding is the core shortfall. Alphabet stock dropped ~4% as three Chinese frontier models launched.
FYIModels ·
GPT-Red Makes GPT-5.6 Sol 6x Harder to Jailbreak
GPT-Red achieved 84% attack success vs human red-teamers' 13%, and adversarially training GPT-5.6 Sol against it made the model 6x more robust to jailbreaks.
FYIModels ·
GPT-5.6 Sol Ultra Confirmed for Codex: GA Imminent
OpenAI's Codex lead confirmed GPT-5.6 Sol Ultra will ship in Codex. Still government-gated, but the signal and markets point to July 9 GA.
Never need to catch up again
The weekly delta: only verdict changes and act-now items. No digest filler.