All stories
Every published Prompt Blueprints story, newest first.
OpenAI says a plan-change error cut Codex allowances on Oct 16, and is adding $500 in credits
An OpenAI Support email, shared on X, says an error kept Pro 200 allowances in place after downgrades. The correction cut weekly allowances abruptly, and OpenAI is adding $500 in credits, separate from the weekly…
Read the featureAll stories
-
Read →Ornith-1.5-122B Released: MIT-Licensed Open Weights, Quantised Builds and Ollama Support
Ornith-1.5-122B is a new open-weight reasoning MoE model under the MIT license with a 262,144-token context. This article covers how to run it with…
-
Read →ChatGPT Memory Study: What the Viral '96%' Figure Actually Measures
A viral X post says ChatGPT secretly profiles every user. The underlying paper audited saved memory entries from 80 accounts and found something…
-
Read →OpenAI Platform Adds a Decisions Playground for Testing the Decisions API
OpenAI's Platform sidebar now shows a Decisions item marked "New" that opens a playground for the Decisions API, per a post on X. OpenAI's guide says…
-
Read →Mercury 3: Haiku-Class Quality at About 4x the Speed, Per Inception's Own Numbers
Inception says its diffusion LLM Mercury 3 matches or beats Claude Haiku 5.5 and GPT-6 Luna on quality while running 4–8x faster. Here is the pricing…
-
Read →GLM-Omni Is a Z.ai Research Model in a New Paper, Not a Confirmed GLM-5.5
A leak post claims Z.ai's GLM-Omni is probably GLM-5.5. The paper names the model but describes a research build on a GLM-4.1V-Thinking base, and…
-
Read →GPT-6.1 Terra Appears in a Leaked Eval Config Screenshot
A screenshot shared by @LuminaBench shows a YAML eval config naming gpt-6.1-terra as the judge model, but neither OpenAI's nor Microsoft's official…
-
Read →China Proposes Export Controls on 23 AI Technologies
China's commerce and science ministries have published a draft adding 23 AI technologies to export controls, with interim registration now and…
-
Read →OpenAI aurora-blue-alpha, green and red model IDs spotted briefly
An outside X account says three OpenAI model IDs, aurora-blue-alpha, aurora-green-alpha and aurora-red-alpha, appeared briefly and were removed.…
-
Read →LL Alpha Free: stealth model appears in OpenCode, maker unnamed
OpenCode lists LL Alpha Free as a free model with maker, context and release unknown. Its data page shows 28B tokens and 1.9K users on Oct 16.
-
Read →Gemini 4 Argon scores 97.5% on ARC-AGI-3, 49.8% on standard harness
ARC Prize's verified ARC-AGI-3 results for Gemini 4 Argon: 97.5% with the Provider Adapter harness, 49.8% with the standard harness, and what the Dec…
-
Read →Ternary Bonsai 2 2B, 4B, 9B: Smallest Files Need PrismML Builds
PrismML's Ternary Bonsai 2 2B, 4B and 9B are out under Apache 2.0. The 0.43–2 GB sizes need PrismML's binaries; stock llama.cpp loads larger Q2_0…
-
Read →Cat Jam Free Is Free on OpenCode, No End Date Announced
OpenCode's Cat Jam Free, a text-only reasoning model with about 262K context, is free on Zen and Go. Its docs note 30-day data retention.
-
Read →Muse Wonder Is Out: Where to Use It and How It Benchmarks
Meta's Muse Wonder is available in the Meta Model API and Muse Code, and Muse Wonder Full is for Muse Code subscribers. Here is who can use each,…
-
Read →Claude Next leak: X post hints at 'purely coded' image output
An X user shares a video he labels 'Claude Next Model' and says upcoming image outputs will be 'purely coded'. Anthropic hasn't announced a model by…
-
Read →Satya Nadella Proposes Treating AI Models as Insider Risks
Satya Nadella's post on X argues that AI models acting inside organisations should be contained, observed and independently checked, like powerful…
-
Read →Fired OpenAI safety researchers say firing chills safety work
Three fired OpenAI safety and alignment researchers say their open letter describes a chilling effect on safety work and on outside collaboration.…
-
Read →SynthID Detector: price, limits and what changed
Google has expanded SynthID Detector globally in English and named OpenAI, NVIDIA and Kakao as partners. It checks for SynthID watermarks, not AI…
-
Read →Gemini 4 Argon surfaces in an unconfirmed X post
A non-company X account, @lyraxana, mentions a "Gemini 4 Argon day" in a greeting. Google has not announced the name, and the post gives no features,…
-
Read →Claude Haiku 5.5 is out, on average about 75% cheaper than Haiku 4.5
Anthropic has released Claude Haiku 5.5, a low-cost small model it says costs about 75% less to run than Haiku 4.5 on average. It adds an adjustable…
-
Read →GPT-6 Codex Reasoning Budget Compared With the API
A community report shows lower reported reasoning-budget values for GPT-6 in Codex than in the API, but does not establish why or whether the figures…
-
Read →Exo Free Listed as a Limited-Time Free Model on OpenCode
Exo Free is listed in OpenCode Zen as a limited-time free model, with reported support for long context, images and tool calls—but with an important…
-
Read →OpenAI outlines a transparency-focused release of AI mathematics
OpenAI says it is releasing mathematical results from an unnamed internal frontier model, along with GitHub materials, Lean formalizations and…
-
Read →OpenAI Reportedly Planned a GitHub Release of Math Solutions
A post attributed to Wired reported that OpenAI planned to publish solutions to hundreds of unsolved math problems on GitHub, but the release remains…
-
Read →Surface Laptop Ultra Specs: Reported RTX Spark Configurations
Zac Bowden reports two Surface Laptop Ultra RTX Spark configurations, 24GB to 128GB memory options, local-AI support and October preorder and…
-
Read →Nano Banana 2.1: Reported Improvements and Availability
An X post announces Nano Banana 2.1 as an image-generation and editing update with reported quality improvements, bug fixes, lower pricing and access…
-
Read →EmbeddingGemma 2: Google’s On-Device Multimodal Model
Google’s EmbeddingGemma 2 maps text, code, images, video and audio into one embedding space for local search and retrieval.
-
Read →DeepSeek V5 Leak: What the Report Actually Says
An X post reports DeepSeek funding, Huawei Ascend infrastructure and a possible V5 launch, but the details remain unconfirmed.
-
Read →Mistral Large 4: API Access and Open-Weights Timeline
Mistral announces Large 4, a natively multimodal model with 1 trillion total parameters, API access today and open weights planned for October.
-
Read →Mistral’s New Model May Be Large 4, but Cyber Claims Are Unclear
A Reuters excerpt shared in an X post says Mistral planned to unveil a model on Oct. 6, while its possible identity and cybersecurity performance…
-
Read →Banbury Road AI’s Kardashev-0.7 Is a 32-Model Swarm
Banbury Road AI describes Kardashev-0.7 as a 32-model trained swarm using RLPS to develop complementary specializations, with reported efficiency…
-
Read →Nano Banana 2.1 Reportedly Appears in Google Flow as “Beluga”
A post reports that Nano Banana 2.1 is live in Google Flow under the codename “beluga.” A supplied screenshot shows the model selected in an…
-
Read →Reflection AI Beam Model: 501B Parameters, 23B Active
Reflection AI has announced Beam, an open-weight model with 501 billion total parameters, 23 billion active parameters and a planned Apache 2.0…
-
Read →OpenRouter Ends Space Bunny Alpha’s Stealth Period
OpenRouter says Space Bunny Alpha’s stealth period has ended after community feedback helped shape the final model, but its identity and reveal…
-
Read →Gemini Notebook Usage Limits Reportedly Change
TestingCatalog reports five-hour refreshes, flexible Notebook-wide usage, estimated artifact costs and a background generation queue.
-
Read →Claude Cowork Tasks Move to Cloud on October 6
Anthropic says new Claude Cowork tasks for Pro and Max users will run in the cloud from October 6, 2026, while tasks already started locally remain…
-
Read →What Is Cohere North 2? Enterprise Features Explained
Cohere presents North 2 as an enterprise agentic AI platform with reusable workflows, workplace connectors, flexible deployment, security controls,…
-
Read →Could Fable 5.5 Be Reaching Claude Code Users?
A report describes Claude Code saving a rolling Fable selection that may route to a newer model, but it does not confirm Fable 5.5 testing or release.
-
Read →Reported ChatGPT Plus Promotion Comes With Regional Uncertainty
An X thread reports a limited-time ChatGPT Plus promotion, but its terms, eligibility, regional availability and official status remain unconfirmed.
-
Read →AI company news roundup: models, privacy and agents
A source-qualified roundup of reported changes involving Gemini, Kolibri-1, Claude, OpenAI, Mistral and Meta, including what remains unconfirmed.
-
Read →What the gemini-eap Google Cloud quota entry means
A report says an entry named `gemini-eap` appeared in Google Cloud quota data, but the observation does not confirm a public Gemini release.
-
Read →Aleph Alpha Kolibri 1: Open Weights and 1M Context
Aleph Alpha’s Kolibri 1 is an Apache 2.0 open-weight MoE model focused on German and English, with long-context support and tool calling.
-
Read →OpenAI release next week? What the X hint actually says
An X post interprets a reply from Tibo as a possible OpenAI release hint, but the evidence does not confirm a model, Astra 6.1 or a DevDay connection.
-
Read →Nano Banana 2.5 “Resplendent Flash” Reported in Image Test
An X post reports that a Nano Banana 2.5 variant called “resplendent flash” appeared in an arena and compares it with Ideogram 4.5 on a demanding…
-
Read →Antigravity Reportedly Adds Claude 5.5 Models as GPT-OSS 120B Removal Looms
An X post and attached screenshots report Claude Opus 5.5 and Sonnet 5.5 availability on paid Antigravity plans, alongside GPT-OSS 120B’s reported…
-
Read →Gemini Notebooks iOS May Reference Nano Banana 2.5 Flash
A single X post reports an internal image-generation identifier in a Gemini Notebooks iOS update, but does not confirm public access to Nano Banana…
-
Read →Antigravity CLI Reportedly Comes to Android Through Termux
An X post reports native Antigravity CLI availability on Android through Termux, but the release and installation command remain independently…
-
Read →What Finances in ChatGPT Can Do and Who Gets Access
The @ChatGPT account says Finances in ChatGPT is rolling out to Free and Go users in the U.S., with reported tools for spending, budgeting, credit,…
-
Read →Xpass Pricing and Plans: Reported Tiers, Features and Unknowns
A reported Xpass bundle combines X, Grok, Cursor and Grok Bot across four tiers, but usage limits, quota rules and launch timing remain unconfirmed.
-
Read →Underdog’s Private Personal AI: What It Announced
Underdog says it is building private, capable AI for consumer devices, but the announcement does not establish a released product, supported hardware…
-
Read →Ling-3.1-flash Is Reported Free on OpenCode With a 262K Context
OpenCode says inclusionAI’s Ling-3.1-flash is free on its platform and lists 560B total parameters, 25B active parameters and a 262K context figure.
-
Read →Grok Reportedly Advised Trump on Capturing Maduro
A Yahoo report says Donald Trump consulted Grok about Venezuelan reaction to Nicolás Maduro’s possible capture and later viewed the chatbot as…
-
Read →Xiaomi MiMo-V2.6 Agent Arena Rankings: Pro vs Flash
Arena reports different strengths for Xiaomi’s MiMo-V2.6-Pro and MiMo-V2.6-Flash, with Pro ranking higher and Flash costing less per task.
-
Read →Claude Sonnet 5 Appears in Google Antigravity Backend Report
An X post and attached screenshot appear to show a Claude Sonnet 5 request in Google Antigravity, but do not confirm public access or official…
-
Read →AI Subscription vs API Cost: What the Multipliers Show
Paweł Huryn’s experiment estimates how much API-priced coding-model usage selected subscriptions provided, while showing why the results depend on…
-
Read →X Calls reportedly launches with links, scheduling and AI backgrounds
A report says X Calls is live at call.x.com, with account-free joining by code or link, scheduling, whiteboard tools, reactions and backgrounds…
-
Read →Fledge Alpha Appears as a Free OpenCode Model, but Its Identity Is Unknown
Fledge Alpha is a free OpenCode listing with encouraging but anecdotal early tests, while its developer, specifications and benchmark record remain…
-
Read →Why Anthropic Reportedly Met Religious Scholars About Claude
A partial New York Times report describes Anthropic’s private meetings with religious scholars and discussions of Claude’s apparent emotions,…
-
Read →GLM-5.3 UNCENSORED EXL3 3.0bpw: What the Release Includes
A source-based explanation of the community-hosted GLM-5.3 UNCENSORED EXL3 quant, its model lineage, reported tests, TabbyAPI setup and tool-calling…
-
Read →Claude Fable 5.5 URL Raises Questions, But Shows No Release
A reported Claude Platform URL references Fable 5.5, but the supplied evidence shows a 404 response and does not confirm an Anthropic model release.
No entries match that filter.