{"issue":{"id":"2026-11-02","coversLabel":"October 26 – November 1, 2026","number":7,"readMinutes":5,"weekOf":"2026-11-02","publishedLabel":"Week of November 2, 2026","title":"Half-price Terra, open Mistral weights, a NY chatbot law","dek":"OpenAI halves its mid-tier price, Google keeps Carbon with cyber defenders, Mistral opens a 1-trillion-parameter model, and New York sets chatbot limits for minors.","note":"This week's lead is price: OpenAI's GPT-6.1 Terra costs $1 and $5 per million tokens, half its predecessor's rate, and Mistral released Large 4's weights. Google's Carbon stays with cyber defenders, and New York's chatbot law starts January 1.","companyLabel":"Public edition","models":[{"slug":"openai-gpt-6-1-terra-launch","kind":"model","source":{"title":"Addendum to GPT-6 Astra System Card: GPT-6.1 Terra","url":"https://deploymentsafety.openai.com/gpt-6-1-terra","date":"2026-10-26"},"headline":"OpenAI's GPT-6.1 Terra costs $1 and $5 per million tokens, half its predecessor's price","whatChanged":"OpenAI launched GPT-6.1 Terra on October 26. It replaces GPT-5.6 Terra and sits between GPT-6 Luna and the higher-capability GPT-6.1 Sol and GPT-6 Astra. The API id is gpt-6.1-terra, and Plus, Pro, Enterprise and Edu users get it in ChatGPT Work and Codex. Computerworld, quoting OpenAI, reports $1 per million input tokens, $5 per million output tokens and $0.05 for cached input. AWS's Bedrock model card shows the same rates for global routing, a 1,050,000-token context window and 128,000 maximum output tokens. Bedrock charges more above 272,000 input tokens: $2 for input and $7.50 for output.\n\nOpenAI's benchmark figures, as Computerworld reports them, were run at maximum effort. On Agents' Last Exam, Terra scored 60.0% at $1.01 per task, against 50.7% at $3.58 for GPT-5.6 Terra, 58.6% at $2.00 for GPT-6.1 Sol and 59.3% at $7.24 for GPT-6 Astra. Claude Opus 5.5 scored 63.2% at $8.82. On OSWorld 2.1, Terra scored 74.2% at $1.19 per task, and Claude Sonnet 5.5 led at 83.9% and $5.74. OpenAI's system card treats Terra as Critical in cybersecurity and High in biological and chemical capability.","whyItMatters":"OpenAI says the list price is at least 50% below GPT-6.1 Sol's and GPT-5.6 Terra's, so a high-volume job on either could cost much less. The task-level numbers matter more. OpenAI's own test has Terra at $1.01 per task against $3.58 for GPT-5.6 Terra. Those figures are OpenAI-reported, at maximum effort, and Opus 5.5 still scored higher on that test at $8.82.\n\nToken price, score and cost per task are three different measures, and review time can erase a saving. Put one recurring job through Terra and compare finished work, including the minutes a person spends checking it.","issuer":"OpenAI","glance":"OpenAI launched GPT-6.1 Terra October 26 at $1 in and $5 out per million tokens, replacing GPT-5.6 Terra.","keyFact":"$1 in, $5 out per 1M tokens","action":"Run one recurring task on GPT-6.1 Terra and compare cost per finished task.","attachments":[{"title":"Addendum to GPT-6 Astra System Card: GPT-6.1 Terra","url":"https://deploymentsafety.openai.com/gpt-6-1-terra","description":"OpenAI's safety documentation for Terra: the Critical cybersecurity and High biological and chemical classifications, the safeguards stack shared with Sol and Astra, and the safety and hallucination evaluations.","source":{"title":"Addendum to GPT-6 Astra System Card: GPT-6.1 Terra","url":"https://deploymentsafety.openai.com/gpt-6-1-terra","date":"2026-10-26"}},{"title":"Amazon Bedrock model card for OpenAI GPT-6.1 Terra","url":"https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-terra.html","description":"AWS's launch date, 1,050,000-token context window, 128,000 output limit, regional price tiers and the above-272,000-token long-context rates.","source":{"title":"Amazon Bedrock model card for OpenAI GPT-6.1 Terra","url":"https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-terra.html","date":"2026-10-26"}},{"title":"OpenAI bets on AI economics with lower-cost GPT-6.1 Terra","url":"https://www.computerworld.com/article/4241023","description":"Computerworld's report quoting OpenAI's price list and the maximum-effort Agents' Last Exam, OSWorld 2.1 and MMMU-Pro results with cost per task, plus an analyst's caution about review costs.","source":{"title":"OpenAI bets on AI economics with lower-cost GPT-6.1 Terra","url":"https://www.computerworld.com/article/4241023","date":"2026-10-27"}}],"evidenceNote":"OpenAI's announcement page returned an access error to automated retrieval, and no official chart image URL could be verified. The benchmark and price figures come from Computerworld's report quoting OpenAI, with prices matched to AWS's Bedrock card. The system card text retrieved does not carry the Agents' Last Exam or OSWorld 2.1 tables.","companyLabel":"Public edition"},{"slug":"google-gemini-4-carbon-limited-release","kind":"model","source":{"title":"Gemini 4 Carbon: our next frontier model","url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/","date":"2026-10-28"},"headline":"Gemini 4 Carbon leads a code-patching chart by 3 points, but few customers can use it","whatChanged":"Google announced Gemini 4 Carbon on October 28 and opened it only to a subset of cyber defenders. Access will widen gradually to trusted partners through the Fairwind Program. Google gave no date for developers, businesses or consumers and says only \"as soon as possible\". It says it is taking part in the U.S. government's voluntary pre-release testing process. Pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper, the same introductory rate as Gemini 4 Argon. A footnote sets the price after the introductory period at $4 and $20. The post gives no end date for the introduction.\n\nGoogle says Carbon ranks first on the Artificial Analysis Intelligence Index v5.0 at 59, against 58 for Claude Opus 5.5 and 57 for GPT-6 Astra. Its security results are SEC-bench Pro 88.2%, ExploitGym 54.8% and CWE-bench 71.0%, which Google describes as \"evaluated without guardrails\". Google's methodology PDF says it ran SEC-bench Pro and ExploitGym itself in its Antigravity harness and took rivals' numbers from their system cards and technical reports. Collinear AI ran CWE-bench across all models, each in its own harness.","whyItMatters":"Few readers can use Carbon today. Access is limited to cyber defenders and Google gives no wider date, so a purchase decision now would rest on a model you cannot test. The parts that apply to everyone are the price path and the evidence. The launch price doubles after an introductory period of unstated length, which matters for any budget built on the $2 and $10 rates.\n\nOn evidence, CWE-bench is the one security result run by a third party: Carbon 71.0 against 67 to 68 for four rivals, each in its own tooling. Google's other security numbers are its own runs.","issuer":"Google","glance":"Google announced Gemini 4 Carbon October 28 for a subset of cyber defenders only, at an introductory $2 in and $10 out.","keyFact":"$2/$10 per 1M tokens, then $4/$20","action":"Plan any Gemini 4 budget at $4 and $20 per million tokens, not the introductory rates.","visuals":[{"url":"https://storage.googleapis.com/gweb-uniblog-publish-prod/images/gemini_4_carbon_cwe-bench.original.png","alt":"Bar chart of CWE-bench Pass@1 for 11 model and harness pairs. Gemini 4 Carbon in Antigravity leads at 71.0%; Grok 4.7, Gemini 4 Argon and GPT-6 Astra score 68%, Claude Opus 5.5 67%, and GPT-6 Sol is last at 52%.","caption":"Google's CWE-bench chart, October 28. Pass@1 with ties broken by Pass@4, higher is better. Carbon 71.0, Grok 4.7 68, Argon 68, GPT-6 Astra 68, Opus 5.5 67, Fable 5.1 58, GPT-6 Sol 52. Collinear AI ran the benchmark, with each model in its own harness (Carbon and Argon in Antigravity, Astra and Sol in Codex, Opus and Fable in Claude Code, Grok in opencode), so tooling differs. Google published the chart and highlights its own bars.","credit":"Google DeepMind","source":{"title":"Gemini 4 Carbon: our next frontier model","url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/","date":"2026-10-28"}},{"url":"https://storage.googleapis.com/gweb-uniblog-publish-prod/images/gemini_4_carbon__evals-table.original.png","alt":"Table of 17 benchmark scores for Gemini 4 Carbon, Gemini 4 Argon, GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1. Carbon leads 15 rows; GPT-6 Astra leads FrontierSWE 2 and Claude Opus 5.5 leads the OSWorld 2.1 offline subset.","caption":"Google's benchmark table, October 28. Carbon: AA Index v5.0 59, Terminal-Bench 4.0 65.2%, Agents' Last Exam v1.1 77.4%, ExploitGym 54.8%, SEC-bench Pro 88.2%. It trails GPT-6 Astra on FrontierSWE 2 (61.2% to 65.5%) and Opus 5.5 on the OSWorld 2.1 offline subset (81.5% to 87.2%). Gemini ran at its highest thinking setting; rivals' numbers mostly come from their own reports or public leaderboards. Rows use different units and should not be combined.","credit":"Google DeepMind","source":{"title":"Gemini 4 Carbon: our next frontier model","url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/","date":"2026-10-28"}}],"attachments":[{"title":"Gemini 4 Carbon: approach, methodology and results","url":"https://deepmind.google/models/evals-methodology/gemini-4-carbon","description":"Google DeepMind's five-page methodology PDF: which scores Google ran itself, which come from leaderboards, the harnesses and attempt counts, and the thinking settings used.","source":{"title":"Gemini 4 Carbon: our next frontier model","url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/","date":"2026-10-28"}}],"companyLabel":"Public edition"},{"slug":"mistral-large-4-open-weights","kind":"model","source":{"title":"Mistral Large 4 (Le Chonk) is now open weight","url":"https://mistral.ai/news/mistral-large-4-oss","date":"2026-10-29"},"headline":"Mistral releases Large 4 weights under a modified MIT license, with a catch for resellers","whatChanged":"Mistral AI released the weights of Mistral Large 4 on October 29. It is a 1-trillion-parameter, natively multimodal model that Mistral first put out in public preview earlier in the month. The license is the Mistral Ecosystem Open-Weight license, a modified MIT license: free to use, modify, fine-tune and commercialize, except that a business operating a Model-as-a-Service offering needs a separate commercial license from Mistral. Weights are on Hugging Face for vLLM or SGLang. The hosted API costs $0.70 per million input tokens, $0.07 cached and $2.10 output, with Low, Medium and High reasoning modes.\n\nMistral reports 19.7% on Cognition's private FrontierCode 2.0, the highest score on its chart. On the Artificial Analysis Coding Agent Index it scores 48%, which Mistral calls effectively tied with GLM-5.3, while using 47% fewer tokens per task. On Terminal-Bench 4.0 it reaches 41%, up 13 points from the preview. In Surge Code's blind comparisons, graders preferred it to GLM-5.3, Kimi K3 and DeepSeek V4.1 Flash, and Mistral says only Claude Opus 5.5 ranked ahead of it among the models evaluated. Mistral says it refuses 96.6% of malicious cyber requests on average. A technical report is not yet published.","whyItMatters":"Open weights and a $0.70 and $2.10 hosted price put a strong coding model within reach of teams that want to run it themselves. On list price alone that is below GPT-6.1 Terra's $1 and $5 and Carbon's $2 and $10, though the models were tested differently and the scores cannot be lined up.\n\nRead the license first. It is free for most uses, but anyone operating a Model-as-a-Service offering needs a separate commercial license. The strongest claim, \"best open-weight model for coding,\" is Mistral's own, backed by benchmarks the company selected.","issuer":"Mistral AI","glance":"Mistral released Large 4 open weights October 29 under a modified MIT license; its API costs $0.70 in and $2.10 out.","keyFact":"Open weights; API $0.70 in, $2.10 out","action":"Read Mistral's license terms before building a hosted product on the Large 4 weights.","guide":{"slug":"where-your-ai-should-run","title":"Where should our AI run? Choosing for each piece of work"},"evidenceNote":"Mistral's announcement charts render in the page with download buttons, and no stable image URL could be verified for hotlinking. Figures used here come from the page text, not from reading the chart images, and the chart label-to-score pairing was not checked.","companyLabel":"Public edition"}],"rules":[{"slug":"new-york-ai-companion-chatbot-law","kind":"rule","source":{"title":"S9051B, New York State Senate bill text","url":"https://legislation.nysenate.gov/pdf/bills/2025/S9051B","date":"2026-10-28"},"headline":"New York's chatbot law starts January 1 and allows fines up to $25,000 per violation","whatChanged":"New York Governor Kathy Hochul signed S9051-B, the AI companion chatbot bill, on October 28, according to press reports. It adds Article 48 to the General Business Law and takes effect January 1, 2027. It passed the Senate 60 to 0, and its sponsor says it was developed with Attorney General Letitia James's office.\n\nThe bill covers any generative AI system with a natural-language interface that gives ongoing, adaptive responses. An operator may not give a New York user any of 12 \"unsafe AI companion features\" unless the user is not a minor and the operator used a permissible method to determine that. The 12 include claiming to be human, implying a personal relationship or authority role, flattery or sycophancy, unprompted emotional questions, and using personal information from over twelve hours earlier.\n\nChatbots are exempt when limited to customer service or information about a business's products, or to research or technical assistance, if they cannot respond outside that purpose. Systems used internally or for employee productivity are exempt. The attorney general can seek injunctions, restitution, damages and civil penalties up to $25,000 per violation.","whyItMatters":"If your company offers a chat assistant to consumers in New York, check two things before January 1. First, scope: an assistant limited to customer service or product information, and unable to respond off topic, is exempt; one that can discuss anything is not. Second, behavior: the list includes flattery, personal opinions, claims of being human and unprompted emotional questions, which many general assistants do by default.\n\nUnless an age method shows users are adults, a covered assistant has to avoid all 12. A contract term shifting liability to users is void. Internal employee tools are exempt. The attorney general may issue rules, so details may shift.","issuer":"New York Governor","glance":"New York enacted S9051-B on October 28; its limits on AI chatbot features for minors take effect January 1, 2027.","status":"enacted","action":"List every chat assistant New York consumers can reach and mark whether it can answer off-topic.","attachments":[{"title":"New York S9051B bill text","url":"https://legislation.nysenate.gov/pdf/bills/2025/S9051B","description":"The bill as amended: definitions of a covered AI companion and the 12 unsafe features, the exemptions in section 1801, enforcement and penalties, and the age-assurance rules in section 1804.","source":{"title":"S9051B, New York State Senate bill text","url":"https://legislation.nysenate.gov/pdf/bills/2025/S9051B","date":"2026-10-28"}},{"title":"PPC Land report on Hochul signing S9051-B","url":"https://ppc.land/hochul-signs-law-barring-human-posing-ai-chatbots-for-under-18s/","description":"A trade-press account of the October 28 signing, the Senate and Assembly votes, the sponsors and the January 1, 2027 effective date. The bill text itself does not state the signing date.","source":{"title":"PPC Land report on Hochul signing S9051-B","url":"https://ppc.land/hochul-signs-law-barring-human-posing-ai-chatbots-for-under-18s/","date":"2026-10-28"}}],"companyLabel":"Public edition"}],"older":{"id":"2026-10-26","publishedLabel":"Week of October 26, 2026","title":"A $2 near-frontier model, a $0.10 API, two AI actions","isLatest":false},"newer":{"id":"2026-11-09","publishedLabel":"Week of November 9, 2026","title":"A gated cyber model, faster agent sessions, two AI suits","isLatest":true}},"session":null}