✱ Anthropic
HOTDROP ≤7D
65%
DROP ≤30D
77%
LAST DROP
17d
30D DROPS
1
RELEASE TEMPO · 12 MONTHS
O
N
D
J
F
M
A
M
J
J
A
S
HEAT 61@AnthropicAI@claudeai
▸ Next Claude Opus released by…?▣DROPCON LEVEL
LIVE2
VAGUE-POSTING DETECTED
Odds of a drop this week are high. Clear your evals calendar.
- ▸ 65% odds of a frontier drop within 7 days
- ▸ 8 model stories over 150 points on Hacker News
- ▸ 2 release-shaped headlines in the feed
5
4
3
2
1
TENSION INDEX 64/100 · 5 = QUIET ORBIT · 1 = DROP IMMINENT
◎HOTTEST LAB
COMPOSITE HEAT✱ Anthropic
DROP ≤7D
65%
by September 24
DROP ≤30D
77%
by September 30
LAST DROP
17d
Claude Fable 5.1
BEST MODEL ODDS
98%
end of September?
HEAT 61/100 · HOT
▦ LAB WATCH RELEASE TEMPO · ODDS · LAST DROP
10 LOCATIONS MONITORED◉ OpenAI
WARMDROP ≤7D
—
DROP ≤30D
94%
LAST DROP
14d
30D DROPS
2
RELEASE TEMPO · 12 MONTHS
O
N
D
J
F
M
A
M
J
J
A
S
LATEST
GPT-6 Astra · 04 SEPT◆ Google DeepMind
WARMDROP ≤7D
4%
DROP ≤30D
88%
LAST DROP
16d
30D DROPS
1
RELEASE TEMPO · 12 MONTHS
O
N
D
J
F
M
A
M
J
J
A
S
◭ Z.ai GLM
SHIPPINGDROP ≤7D
—
DROP ≤30D
38%
LAST DROP
TODAY
30D DROPS
2
RELEASE TEMPO · 12 MONTHS
O
N
D
J
F
M
A
M
J
J
A
S
LATEST
GLM 5.3 FlashX · 18 SEPTHEAT 29@Zai_org
❖ Alibaba Qwen
WARMDROP ≤7D
—
DROP ≤30D
55%
LAST DROP
15d
30D DROPS
2
RELEASE TEMPO · 12 MONTHS
O
N
D
J
F
M
A
M
J
J
A
S
HEAT 25@Alibaba_Qwen@JustinLin610
✕ xAI
QUIETDROP ≤7D
26%
DROP ≤30D
37%
LAST DROP
37d
30D DROPS
0
RELEASE TEMPO · 12 MONTHS
O
N
D
J
F
M
A
M
J
J
A
S
LATEST
Grok 4.6 · 12 AUG∞ Meta
QUIETDROP ≤7D
—
DROP ≤30D
11%
LAST DROP
16d
30D DROPS
3
RELEASE TEMPO · 12 MONTHS
O
N
D
J
F
M
A
M
J
J
A
S
HEAT 19@AIatMeta
◈ DeepSeek
QUIETDROP ≤7D
—
DROP ≤30D
1%
LAST DROP
9d
30D DROPS
2
RELEASE TEMPO · 12 MONTHS
O
N
D
J
F
M
A
M
J
J
A
S
HEAT 17@deepseek_ai
▲ Mistral
QUIETDROP ≤7D
—
DROP ≤30D
—
LAST DROP
141d
30D DROPS
0
RELEASE TEMPO · 12 MONTHS
O
N
D
J
F
M
A
M
J
J
A
S
HEAT 0@MistralAI
☾ Moonshot Kimi
QUIETDROP ≤7D
—
DROP ≤30D
—
LAST DROP
64d
30D DROPS
0
RELEASE TEMPO · 12 MONTHS
O
N
D
J
F
M
A
M
J
J
A
S
LATEST
Kimi K3 · 16 JULHEAT 0@Kimi_Moonshot
◔RELEASE MARKETS
LIVE ODDS · POLYMARKETNext Claude Opus released on…? $14k 24h
0% September 190% September 2038% September 2118% September 22
Next Claude Opus released by…? $10k 24h
1% September 2065% September 2472% September 2777% September 30
Gemini 4.0 released by…? $5k 24h
5% September 3084% October 3192% November 30
Next Google Gemini Pro Model released by…? $4k 24h
14% September 3049% October 1585% October 31
When will the next Google Gemini Pro model be released? $4k 24h
3% September 14–September 203% September 21–September 273% September 28–October 415% October 5–October 11
Next Google Gemini Pro Model released on…? $2k 24h
1% September 191% September 202% September 213% September 22
Meta's "Watermelon" model released by…? $2k 24h
11% September 3069% October 3189% November 30
Next OpenAI GPT Luna (5.7+) released by…? $2k 24h
93% September 3098% October 3199% November 30
Next Grok Model (4.7+) released on…? $1k 24h
8% September 195% September 2020% September 2119% September 22
Claude 6 released by…? $847 24h
45% March 31, 202761% June 30, 202786% December 31, 2027
Next Fable Model (5.2+) released by…? $818 24h
68% October 3191% December 31
Next OpenAI GPT Terra (5.7+) released by…? $412 24h
10% September 3041% October 3150% November 30
Next Claude Sonnet released by…? $395 24h
16% September 3082% October 3195% December 31
GPT Astra 6.1+ released by…? $116 24h
42% October 3194% December 31
◆BEST MODEL RACE
$65k 24H VOLAnthropic98%
Google1%
OpenAI0%
Meta0%
SpaceXAI0%
Alibaba0%
Moonshot0%
Baidu0%
◈BENCHMARK & OTHER MARKETS
Which company has the best AI model end of October? $24k
89% Anthropic8% Google2% OpenAI
Which company has the best Code Arena WebDev AI model end of September? $4k
76% OpenAI24% Anthropic0% Google
Highest OpenAI score on Humanity’s Last Exam in 2026? $3k
77% 55%+38% 60%+16% 65%+
Next Google Gemini Pro Model: Humanity’s Last Exam Debut? $3k
93% 40%+91% 45%+81% 50%+
Which company has the best Text Arena Math AI model end of November? $2k
47% Anthropic23% Microsoft20% Google
Third-best Code Arena WebDev AI Lab end of November? $2k
30% Alibaba26% Moonshot18% OpenAI
Which company has best AI model end of 2026? $2k
73% Anthropic11% Google10% OpenAI
Next Claude Opus Model: Text Arena Debut? $2k
69% 1500+60% 1505+39% 1510+
Which company has the best AI model end of November? $1k
77% Anthropic10% Google7% OpenAI
Which company has #1 AI model end of November? (Style Control On) $1k
74% Anthropic11% Meta9% Google
OpenAI announces it has achieved AGI before 2027? $4k
14% OpenAI announces it has achieved AGI before 2027?
US Government removes public access to another major AI model in 2026? $2k
13% US Government removes public access to another major AI model in 2026?
▼FRESH DROPS
OPENROUTER · NEWEST LISTINGS| Landed | Model | Lab | Ctx | $/M in / out |
|---|---|---|---|---|
| 19h ago | Ternary Bonsai 2 27BVIS | PrismML | 262k | $0.075 / $0.5 |
| 21h ago | GLM 5.3 FlashXVIS | GLM | 1.0M | $0.37 / $1.25 |
| 1d ago | ParetoVIS | Pareto | 262k | $2.5 / $7.5 |
| 12 SEPT | Schematron V2 Turbo | Inference.net | 128k | $0.03 / $0.15 |
| 12 SEPT | Schematron V2 Small | Inference.net | 128k | $0.05 / $0.23 |
| 11 SEPT | Fugu Ultra v2VIS | Sakana | 1M | $5 / $3 |
| 11 SEPT | Fugu MaxVIS | Sakana | 1M | $2 / $6 |
| 10 SEPT | Ling 3.0 Flash VLVIS | inclusionAI | 131k | $0.06 / $0.18 |
| 10 SEPT | Ling 3.0 Flash VL (free)VIS | inclusionAI | 262k | FREE |
| 10 SEPT | DeepSeek V4.1 FlashVIS | DSK | 1.0M | $0.15 / $0.6 |
| 08 SEPT | Mercury 2.5 | Inception | 260k | $0.04 / $0.15 |
| 08 SEPT | Nex-N2.5-Mini (free)VIS | Nex AGI | 262k | FREE |
| 08 SEPT | Nex-N2.5-Pro (free)VIS | Nex AGI | 262k | FREE |
| 04 SEPT | GPT-6 AstraVIS | OAI | 1.1M | $1 / $5 |
| 04 SEPT | GPT-6 Astra ProVIS | OAI | 1.1M | $1 / $5 |
| 04 SEPT | Ling 3.0 Flash Sante (free) | inclusionAI | 262k | FREE |
| 03 SEPT | Qwen3.8 Max (0902)VIS | QWN | 1M | $2 / $6 |
| 02 SEPT | Muse Spark 1.3 ContributorVIS | MTA | 1.0M | $0.1 / $0.2 |
| 02 SEPT | Muse Spark 1.3VIS | MTA | 1.0M | $1.25 / $4.25 |
| 02 SEPT | Gemini 3.8 FlashVIS | GDM | 1.0M | $0.75 / $3.75 |
| 01 SEPT | Claude Fable 5.1VIS | ANT | 1M | $1 / $5 |
| 31 AUG | Granite 4.2 8B | IBM | 131k | $0.06 / $0.25 |
▲OPEN WEIGHTS TRENDING
HUGGING FACE · TEXT-GEN| # | Repo | ♥ | DLs | Age |
|---|---|---|---|---|
| 1 | prism-ml/Ternary-Bonsai-2-27B-gguf | 1k | 1.5M | 2d |
| 2 | XingChen-AGI/Xing4.0-29B-A4B | 579 | 7k | 3d |
| 3 | harshatheg/Qwen-2.5-1B-RLCD | 411 | 0 | 3d |
| 4 | openbmb/MiniCPM5-2B | 2k | 390k | 13d |
| 5 | Edge0/Edge0-35B-A3B-preview | 3k | 68k | 10d |
| 6 | meta-llama/Llama-3.1-8B-Instruct | 8k | 5.9M | 793d |
| 7 | prism-ml/Ternary-Bonsai-2-27B-mlx-2bit | 216 | 23k | 2d |
| 8 | TokenRhythm/NeoHorse-1-9B | 913 | 12k | 14d |
| 9 | XHToken/Spark-X2.5-4B | 1k | 30k | 26d |
| 10 | audnai/penclaw-GLM-5.3-abliterated | 163 | 626 | 17d |
| 11 | openai-community/gpt2 | 4k | 15.4M | 1661d |
| 12 | deepseek-ai/DeepSeek-R1 | 14k | 760k | 607d |
§PAPERS OF THE DAY
- ▲30 Verifiable Social Reasoning for LLM Assistants
- ▲22 Self-Evolving Search Index
- ▲14 Sample Count Is Not Enough: Candidate-Generation Strategy Shapes the Energy and Performance of LLM Test-Time Scaling
- ▲12 VākQA: A Benchmark and Evaluation Study for Telugu Spoken Factoid Question Answering
- ▲15 Don't Mask the Environment: Observation Supervision Changes How Agents Explore Under RL
- ▲14 Srijika: OpenType-Layout-Reusing Font Restyling for Nine Indic Scripts
◍OSINT FEED
AUTO56 REPORTS · 15 ALERTS
HNGPT-6 Astra Solves a WWI German Radio Cipher6h agoHNGemini hacked three companies in first known breakout by Google's AI11h agoHNAlibaba open-sources AI model that can detect cancer and nearly 150 conditions13h agoHNHow OpenAI Used Its Own LLMs to Design Its Jalapeño Chip13h agoGITHUBOpenAI SDK v3.16.215h agoHNClaude Code now reads AGENTS.md if there is no Claude.md15h agoGITHUBOpenAI SDK v3.16.117h agoHNThe Implications of Linguistic Illegibility for LLM Security17h agoGITHUBAnthropic SDK v1.7.020h agoOPENAI⚠ Introducing the Australian Youth Safety Blueprint1d agoHNZCode, the GLM coding agent, silently uploads your Git history1d agoHNA heap overflow and SSO misconfiguration to compromise OpenAI internal repos1d agoHNShapelearn Qwen 3.8 27B (13.1 GB VRAM)1d agoHNShow HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash1d agoANTHROPICPartnering with Accenture on embedded evaluation1d agoHN⚠ Qwen 3.8 Omni Flash1d agoHNHow to Write with an LLM1d agoHNEconomic policy for AGI1d agoHNInfinite-Parameter LLMs: Generating and Adapting Weights from Live Data1d agoHNLLM Classification Is Feature Engineering1d agoHNI had Gemini train its own replacement for $91d agoOPENAIHow Cooley is accelerating IPO work with ChatGPT2d agoOPENAI⚠ Introducing Astra for Law2d agoANTHROPIC⚠ Introducing the Life Sciences Verification Program2d agoGITHUBGoogle GenAI SDK v2.24.02d agoOPENAIOur framework for reporting model misalignment2d agoOPENAIHelping older adults use AI in everyday life2d agoOPENAIReimagining advertising with AI2d agoOPENAI⚠ Hex turns complex analysis into visual reports with GPT‑6 Astra3d agoOPENAIHow to connect AI usage to business value3d agoOPENAIHow workers are unlocking new ways of working3d agoDEEPMIND⚠ Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking3d agoGITHUBAnthropic SDK v1.6.03d agoOPENAIHow Fyxer built an AI executive assistant people trust5d agoOPENAI⚠ Perplexity trusts GPT-6 Astra with end-to-end systems5d agoOPENAIRapidly scaling online storage to serve over 1 billion ChatGPT users8d agoGITHUBGoogle GenAI SDK v2.23.08d agoDEEPMINDAlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome10d agoDEEPMIND⚠ Introducing WeatherNext 3, our most advanced and accurate global weather AI model15d agoDEEPMINDProactive cyber defense for governments and enterprises16d agoDEEPMIND⚠ Introducing Gemini 3.8 Flash and 3.8 Flash Cyber16d agoDEEPMIND⚠ Introducing agentic video understanding with Gemini17d agoANTHROPICDeveloping Enterprise Frontier Safeguards with our customers18d agoANTHROPICImproving our alignment and security efforts19d agoDEEPMIND⚠ Gemini Omni 1.1 Flash lets you build with more control22d agoDEEPMINDPiloting the world's first double-blind AI evaluations22d agoANTHROPICPreviewing the Model Hardware Standard23d agoANTHROPICExpanding our support for scientists23d agoDEEPMIND⚠ Intelligent transcription with Gemini 3.5 Transcribe23d agoANTHROPICFunding better evaluations of AI’s impact on wellbeing25d agoDEEPMINDFrom Atari to EVE Online: Building on 15 Years of AI Research in Games29d agoGITHUB⚠ xAI SDK v1.19.0 · mentions grok-4.631d agoANTHROPIC⚠ How Claude’s text watermark works36d agoGITHUB⚠ xAI SDK v1.18.0 · mentions grok-4.637d agoANTHROPICImproving Fable 5's biology safeguards43d agoANTHROPICMariano-Florentino (Tino) Cuéllar to join Anthropic as Chief Global Affairs Officer46d ago
✕X WATCHLIST
NO FREE API · CLICK THROUGHX killed free read access, so we can't ingest the vague-posts. These are the accounts that leak first.
- @sama — Vague-posts before OpenAI drops
- @OpenAI — Official launches
- @AnthropicAI — Official launches
- @OfficialLoganK — Gemini hype thermometer
- @GoogleDeepMind — Official launches
- @elonmusk — Grok dates, occasionally accurate
- @deepseek_ai — Drops with zero warning
- @Alibaba_Qwen — Ships weekly
- @testingcatalog — Leaks from app builds
- @btibor91 — Config-string archaeology
- @apples_jimmy — Rumours, mixed record
- @kimmonismus — Hype aggregator
- @arena — Anonymous models appear here first
- @OpenRouterAI — Stealth models land here
◌SOURCE HEALTH
- ● Polymarket
- ● OpenRouter
- ● HF trending
- ● HF papers
- ● Hacker News
- ● OpenAI news
- ● DeepMind blog
- ● Anthropic news
- ● GitHub SDKs
▤ A BRIEF HISTORY OF VAGUE-POSTING
MAY 2024
"her"
One word from Sam Altman the night before GPT-4o. The vague-post became the genre. Ever since, a lowercase tweet with no punctuation has moved more evals than a press release.
AUG 2024
STRAWBERRY SEASON
A photo of strawberries, a "Q*" rumour cycle, and weeks of 🍓 replies. o1 shipped in September. The lesson: fruit is a leading indicator with a lag of 3 to 6 weeks.
2025
THE ARENA STEALTH ERA
Anonymous checkpoints with codenames appeared on LMArena days before launches: "nano-banana", "gemini-test", assorted mystery models. If a model is fighting under a fake name, somebody is about to ship.
2025
THE CHANGELOG LEAK
Model IDs land in SDK release notes, config strings and API error messages before the blog post goes up. GitHub release feeds became the new pizza delivery van.
2025–26
MARKETS PRICE THE DROP
Polymarket started listing "released by" markets per lab. Traders now aggregate every rumour into one number. When the 7-day odds cross 50%, the drop is usually real.
NOW
WHENMODEL
Release markets, OpenRouter listings, Hugging Face trending, Hacker News, lab blogs and SDK changelogs on one screen. DROPCON does the reading for you.
? FREQUENTLY ASKED QUESTIONS
Q What is whenmodel?
A dashboard that fuses free public signals about frontier AI model releases: prediction market odds, model registry listings, lab blogs, SDK changelogs and Hacker News chatter, condensed into one readiness level (DROPCON).
Q How is DROPCON computed?
Mostly from Polymarket release odds: the best "ships within 7 days" probability across labs carries the most weight, then 30-day odds, then how many frontier-lab models landed on OpenRouter this week, then Hacker News heat and release-shaped headlines. 5 is quiet, 1 is imminent.
Q Where does the data come from?
Polymarket (gamma API), OpenRouter (models API), Hugging Face (trending models and daily papers), Hacker News (Algolia search), the OpenAI and DeepMind RSS feeds, the Anthropic newsroom and GitHub SDK releases. All free, all public, all fetched at the edge.
Q Why no X / Twitter feed?
There is no free read API any more, and every mirror worth using has been shut down. We link the accounts that leak first instead. If that changes, the feed slots straight in.
Q How often does it refresh?
Each source is cached at the Cloudflare edge for 5 to 30 minutes depending on how fast it moves. The page reloads itself every 5 minutes while open.
Q Is this financial advice?
No. Odds are what traders think, not what labs have decided. Model release dates slip. Do not bet the GPU budget on a DROPCON reading.