{"issue":{"id":"2026-10-26","coversLabel":"October 19–25, 2026","number":6,"readMinutes":5,"weekOf":"2026-10-26","publishedLabel":"Week of October 26, 2026","title":"A $2 near-frontier model, a $0.10 API, two AI actions","dek":"Moonshot's Kimi K3.1 Pro and OpenAI's Decisions API arrive with new prices, Texas sues OpenAI over what its ads left out, and SEC examiners are told to watch AI phishing and deepfakes.","companyLabel":"Public edition","models":[{"slug":"moonshot-kimi-k3-1-pro-launch","kind":"model","source":{"title":"Introducing Kimi K3.1 Pro","url":"https://www.kimi.ai/blog/kimi-k3-1-pro","date":"2026-10-23"},"headline":"Kimi K3.1 Pro costs $2 and $10 per million tokens, with weights promised by Nov 4","whatChanged":"Moonshot AI released Kimi K3.1 Pro on October 23 through Kimi.ai, Kimi Work, Kimi Code and the Kimi API, with a 1 million token context window. API pricing is $2 per million input tokens, $10 per million output tokens and $0.20 for cached input, one-third below the $3 and $15 Moonshot lists for Kimi K3. A Highspeed variant, which Moonshot says is 2 to 3 times faster, costs $4 and $20. Moonshot promises full model weights by November 4. They are not out yet, and Artificial Analysis currently labels the model proprietary.\n\nMoonshot's chart compares it with Claude Opus 5.5, GPT-6 Astra and Claude Fable 5.1. On Terminal-Bench 4.0 Kimi scores 65.7 against 66.4 for Opus 5.5 and 58.2 for Astra. It leads on SWE Marathon (51.7 against 46.3 for Opus 5.5) and OfficeQA Pro v2 (64.4 against 62.2). It trails on Terminal-Bench Science 0.1 (38.1 against 68.1 for Astra) and Frontier Finance (58.6 against 60.8 for Fable 5.1). Moonshot ran Kimi at maximum reasoning effort and took rival scores from their blogs or public leaderboards.\n\nArtificial Analysis, an independent tester, scores Kimi 55 on its Intelligence Index at $2.11 per index task. The model produced 150 million output tokens on the index against a median of 81 million for similar models, at 46.3 tokens per second against a median of 87 for similarly priced models.","whyItMatters":"The token price is only part of the cost of a task. Artificial Analysis measured 150 million output tokens for its index against a median of 81 million, and a speed of 46.3 tokens per second. A job billed at $10 per million output tokens can cost more and run slower than the rate card suggests.\n\nMoonshot's scores are vendor-run, and the chart mixes its own harness with rivals' published numbers. Several rows differ by 2 points or less. Weights are promised but not released, so self-hosting is a plan and not an option today. Run one recurring task of your own and compare cost per finished task.","issuer":"Moonshot AI","glance":"Moonshot released Kimi K3.1 Pro October 23 at $2 in and $10 out per million tokens; open weights are promised by November 4.","keyFact":"$2 in, $10 out per 1M tokens; weights by Nov 4","action":"Run one recurring task on Kimi K3.1 Pro and compare cost per finished task with your current model.","visuals":[{"url":"https://kimi-img.kimi.ai/pub/slides/kimi-fs/managed-assets/6f2d669b-c9e2-4579-bf7c-26a0b78ced74.webp","alt":"Grid of horizontal bar charts comparing Kimi K3.1 Pro with Claude Opus 5.5, Claude Fable 5.1, GPT-6 Astra, DeepSeek-V4.1-Flash-1017 and GLM-5.3. Kimi leads SWE Marathon and OfficeQA Pro v2 and trails on Terminal-Bench Science 0.1.","caption":"Moonshot's benchmark grid, October 23. Terminal-Bench 4.0: Opus 5.5 66.4, Kimi 65.7, GPT-6 Astra 58.2. SWE Marathon: Kimi 51.7, Opus 5.5 46.3. OfficeQA Pro v2: Kimi 64.4, Opus 5.5 62.2. Frontier Finance: Fable 5.1 60.8, Kimi 58.6. Terminal-Bench Science 0.1: Astra 68.1, Kimi 38.1. Vendor-run at maximum reasoning effort; rival scores are Moonshot's picks from their blogs or leaderboards, and the grid includes Moonshot's own Kimi Code Bench. Rows differ in units, so scores should not be combined.","credit":"Moonshot AI","source":{"title":"Introducing Kimi K3.1 Pro","url":"https://www.kimi.ai/blog/kimi-k3-1-pro","date":"2026-10-23"}}],"attachments":[{"title":"Kimi K3.1 Pro pricing, standard and Highspeed","url":"https://www.kimi.ai/resources/kimi-k3-1-pro-pricing","description":"Per-token rates for kimi-k3.1-pro and the Highspeed model, cache-write prices, the flat rate across the 1,048,576-token window, and the list prices for Kimi K3.","source":{"title":"Introducing Kimi K3.1 Pro","url":"https://www.kimi.ai/blog/kimi-k3-1-pro","date":"2026-10-23"}}],"companyLabel":"Public edition"},{"slug":"openai-decisions-api-general-availability","kind":"model","source":{"title":"Build more responsive apps with the Decisions API","url":"https://openai.com/index/decisions-api/","date":"2026-10-22"},"headline":"OpenAI charges $0.10 per million input tokens and nothing for output on Decisions","whatChanged":"OpenAI made the Decisions API generally available on October 22. It runs on GPT-6 Luna, the only supported model, through a dedicated endpoint. A request supplies text, images or JSON plus up to 400 questions, double the beta limit. Each question returns one of three answer types: the probability that a condition is true, a choice from supplied options with probabilities and a confidence value, or a score that averages ordered rubric levels by probability. OpenAI's developer guide points to the Responses API when a written explanation is needed.\n\nUncached input costs $0.10 per million tokens and cached input $0.01. OpenAI charges nothing for output tokens. The guide says regional processing premiums and long-context multipliers apply. OpenAI says the API is up to 10 times faster than GPT-6 Luna through the Responses API. That is OpenAI's claim, and the announcement publishes no workload, accuracy result or Responses API price to check it against.\n\nSince the beta, OpenAI added prompt caching, inference residency in the US and EU, native JSON input, images in question instructions and a GPT-Live integration. Zero Data Retention and HIPAA use are available to eligible customers.","whyItMatters":"This is a price for bounded questions: routing a support ticket, tagging a document, or deciding whether an agent should retry. With no output charge, a classification job costs roughly in proportion to what you send, and one shared document can be asked many questions in a single request.\n\nThe number to test is accuracy. OpenAI publishes none, and a probability of 0.9 is only useful if it holds on your own labeled examples. The developer guide says to set thresholds from them. Check the regional-processing premium before comparing the price with your current bill.","issuer":"OpenAI","glance":"OpenAI made its Decisions API generally available October 22 at $0.10 per million input tokens, with no output-token charge.","keyFact":"$0.10 per 1M input tokens; no output charge","action":"Test Decisions on 50 labeled examples from one routing task before comparing cost.","attachments":[{"title":"Decisions API developer guide","url":"https://developers.openai.com/api/docs/guides/decisions","description":"Request format, the three question types, the 1,024-token caching threshold, the pricing and availability section with regional premiums, and the advice to set thresholds from labeled examples.","source":{"title":"Build more responsive apps with the Decisions API","url":"https://openai.com/index/decisions-api/","date":"2026-10-22"}}],"evidenceNote":"OpenAI's announcement has an interactive demo and an answer-type table but no benchmark, accuracy result or pricing comparison, and no Responses API baseline price. No relevant visual evidence could be verified.","companyLabel":"Public edition"}],"rules":[{"slug":"texas-sues-openai-and-altman","kind":"rule","source":{"title":"Texas's Original Petition, State of Texas v. OpenAI Global, LLC et al.","url":"https://www.texasattorneygeneral.gov/sites/default/files/images/press/OpenAI%20Final.pdf","date":"2026-10-23"},"headline":"Texas sues OpenAI, saying ads for ChatGPT left out that it can be wrong","whatChanged":"On October 23, Texas Attorney General Ken Paxton announced a suit against OpenAI Global, LLC, five affiliated OpenAI entities and CEO Sam Altman personally. The petition was filed in a Collin County district court, and the copy the Attorney General published has no cause number. It pleads the Texas Deceptive Trade Practices-Consumer Protection Act and the SCOPE Act (Securing Children Online through Parental Empowerment), citing sections 509.051, 509.052, 509.054 and 509.101. Texas seeks monetary relief over $1,000,000 plus injunctions, and asks the court to hold Altman liable for his direct involvement.\n\nThe deception theory rests on statements and omissions. The petition says OpenAI's advertising shows ChatGPT running a farm and handling health questions without disclosing that it can be wrong or make things up (paragraph 54). The SCOPE Act counts rest on allegations that the free version has no age verification and that parents lack the controls the Act requires for known minors (paragraphs 74 to 77).\n\nThe press release's list of harms is the State's allegation. Nothing has been decided, and the State's materials include no OpenAI response.","whyItMatters":"The theory needs no AI statute. A company that says its AI is safe, accurate or reliable, and omits that it can be wrong, can be pleaded as deceptive under a general consumer-protection law. The petition quotes OpenAI's social posts, a farm-management campaign and blog text.\n\nAny company selling AI features to Texas consumers could read its own ad copy, help text and onboarding warnings against what the product does. Companies whose tools reach minors should also look at age checks and parental controls, the subject of the SCOPE Act counts. A company that only buys ChatGPT is not a defendant. Claims it repeats to its own customers are its own.","issuer":"Texas Attorney General","glance":"Texas sued OpenAI and Sam Altman on October 23 under its consumer-protection and SCOPE Acts; the claims are allegations.","status":"proposed","action":"Read your AI feature ads and help text for accuracy or safety claims, and note the proof for each.","attachments":[{"title":"Texas's Original Petition against OpenAI and Sam Altman","url":"https://www.texasattorneygeneral.gov/sites/default/files/images/press/OpenAI%20Final.pdf","description":"The filed petition: the parties, the DTPA and SCOPE Act sections pleaded, the $1,000,000 relief statement, and the quoted advertising the State says omitted error risk.","source":{"title":"Texas's Original Petition, State of Texas v. OpenAI Global, LLC et al.","url":"https://www.texasattorneygeneral.gov/sites/default/files/images/press/OpenAI%20Final.pdf","date":"2026-10-23"}},{"title":"Texas Attorney General press release","url":"https://www.texasattorneygeneral.gov/news/releases/attorney-general-ken-paxton-sues-openai-and-ceo-sam-altman-lying-users-and-harming-texans","description":"The State's summary of the suit and its list of alleged harms. It names only the DTPA claim; the SCOPE Act claim appears in the petition.","source":{"title":"Attorney General Ken Paxton Sues OpenAI and CEO Sam Altman for Lying to Users and Harming Texans","url":"https://www.texasattorneygeneral.gov/news/releases/attorney-general-ken-paxton-sues-openai-and-ceo-sam-altman-lying-users-and-harming-texans","date":"2026-10-23"}}],"companyLabel":"Public edition"},{"slug":"sec-fy2027-examination-priorities","kind":"rule","source":{"title":"SEC Announces Fiscal Year 2027 Examination Priorities","url":"https://www.sec.gov/newsroom/press-releases/2026-109-sec-announces-fiscal-year-2027-examination-priorities","date":"2026-10-23"},"headline":"SEC examiners are told to look at AI phishing, deepfakes and customer-data safeguards","whatChanged":"On October 23 the SEC's Division of Examinations published its fiscal year 2027 examination priorities (press release 2026-109). The document is a staff statement. Its disclaimer says it is not a rule, has no legal force and creates no new obligations. The division covers more than 34,000 registrants. It uses the term \"super intelligence (SI)\" for the tools it describes and says registrants use \"generative SI tools to support research, drafting, coding, and client communications\" as well as advisory, portfolio management and compliance monitoring. It says it is intensifying its efforts to see whether firms comply with existing, technology-neutral obligations.\n\nExaminers will assess preparation for SI-related cyberattacks, such as SI-generated phishing and deepfake voice and video used for social engineering. They will prioritize policies for monitoring SI in fraud prevention, back-office operations, anti-money laundering, Regulation Best Interest compliance and trading. For automated investment tools they will ask whether representations are fair and accurate.\n\nUnder the SEC's customer-information safeguards rule, examiners will review incident response programs, including customer notification and oversight of service providers. Under its identity theft red-flags rule, they will check whether programs address risks tied to SI and stolen credentials. The division says the list is not exhaustive.","whyItMatters":"Examiners are told to look at how firms use AI and at how attackers use it. For a regulated firm, or a vendor to one, that is a concrete list: AI-written phishing, synthetic voice and video used for social engineering, and safeguards for customer information. The document creates no new duty.\n\nAny organization that holds customer data can use the same list. Find out which AI tools touch that data, who approved them, and whether staff know how to verify an unexpected request that arrives in a familiar voice.","issuer":"SEC Division of Examinations","glance":"SEC examiners' fiscal 2027 priorities, out October 23, cover AI use and AI-enabled attacks; the SEC's wording is \"super intelligence.\"","status":"guidance","action":"List every AI tool that touches customer data and who approved it.","attachments":[{"title":"Fiscal Year 2027 Examination Priorities","url":"https://www.sec.gov/files/2027-exam-priorities.pdf","description":"The full staff document: its no-legal-force disclaimer, the SI passages in the leadership message, the Emerging Financial Technology section, and the sections on customer information, identity theft and systems compliance.","source":{"title":"SEC Announces Fiscal Year 2027 Examination Priorities","url":"https://www.sec.gov/newsroom/press-releases/2026-109-sec-announces-fiscal-year-2027-examination-priorities","date":"2026-10-23"}}],"companyLabel":"Public edition"}],"older":{"id":"2026-10-19","publishedLabel":"Week of October 19, 2026","title":"Three new models, a price clock, two AI vendor probes","isLatest":false},"newer":{"id":"2026-11-02","publishedLabel":"Week of November 2, 2026","title":"Half-price Terra, open Mistral weights, a NY chatbot law","isLatest":false}},"session":null}