<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Oxelan Signal — #inference</title>
    <link>https://blog.oxelan.com</link>
    <description>Posts tagged inference on AI, security and engineering.</description>
    <language>en-us</language>
    <lastBuildDate>Fri, 20 Nov 2026 19:39:43 GMT</lastBuildDate>
    <managingEditor>hello@oxelan.com (Kusal Tharindu)</managingEditor>
    <copyright>© 2026 Oxelan</copyright>
    <atom:link href="https://blog.oxelan.com/tags/inference/rss.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>OpenAI&apos;s MedChemBench puts medicinal chemistry reasoning to the test</title>
      <link>https://blog.oxelan.com/ai/openai-s-medchembench-puts-medicinal-chemistry-reasoning-to/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/openai-s-medchembench-puts-medicinal-chemistry-reasoning-to/</guid>
      <pubDate>Fri, 20 Nov 2026 19:39:43 GMT</pubDate>
      <description>OpenAI introduced MedChemBench, a benchmark of 99 problems drawn from published medicinal chemistry research, and published scores for several GPT models at the xhigh reasoning setting.</description>
      <category>inference</category>
      <category>ai-governance</category>
    </item>
    <item>
      <title>Gemini 4 Neon arrives with top coding-agent scores and $3 input pricing</title>
      <link>https://blog.oxelan.com/ai/gemini-4-neon-arrives-with-top-coding-agent-scores-and-3/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/gemini-4-neon-arrives-with-top-coding-agent-scores-and-3/</guid>
      <pubDate>Fri, 20 Nov 2026 05:56:43 GMT</pubDate>
      <description>Gemini 4 Neon ranks first on the Artificial Analysis Coding Agent Index and is priced at $3 per million input tokens with a 1 million-token context window.</description>
      <category>inference</category>
      <category>llm-security</category>
      <category>ai-agents</category>
    </item>
    <item>
      <title>Anthropic prices its initial public offering at $83.00 per share</title>
      <link>https://blog.oxelan.com/ai/anthropic-prices-its-initial-public-offering-at-83-00-per/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/anthropic-prices-its-initial-public-offering-at-83-00-per/</guid>
      <pubDate>Fri, 20 Nov 2026 05:09:10 GMT</pubDate>
      <description>Anthropic priced its initial public offering of Class A common stock at $83.00 per share, and the shares are expected to begin trading on the Nasdaq Global Select Market on November 19, 2026.</description>
      <category>ai-governance</category>
      <category>inference</category>
    </item>
    <item>
      <title>Palo Alto Networks to buy Applied Compute for $4.5 billion</title>
      <link>https://blog.oxelan.com/security/palo-alto-networks-to-buy-applied-compute-for-4-5-billion/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/security/palo-alto-networks-to-buy-applied-compute-for-4-5-billion/</guid>
      <pubDate>Thu, 19 Nov 2026 03:38:12 GMT</pubDate>
      <description>Palo Alto Networks agreed to acquire AI infrastructure company Applied Compute in an all-stock deal valued at $4.5 billion, and plans to offer its platform to security customers.</description>
      <category>open-weights</category>
      <category>inference</category>
    </item>
    <item>
      <title>Genie 4 arrives in Project Genie with mid-session prompting</title>
      <link>https://blog.oxelan.com/ai/genie-4-arrives-in-project-genie-with-mid-session-prompting/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/genie-4-arrives-in-project-genie-with-mid-session-prompting/</guid>
      <pubDate>Wed, 18 Nov 2026 18:42:04 GMT</pubDate>
      <description>Project Genie now gives first-time access to Genie 4, which adds mid-session prompting and real-time native audio, while Genie 3 worlds gain longer sessions and panorama seeding.</description>
      <category>ai-agents</category>
      <category>inference</category>
    </item>
    <item>
      <title>OpenAI used frontier models to help design its Jalapeño inference chip</title>
      <link>https://blog.oxelan.com/ai/openai-used-frontier-models-to-help-design-its-jalapeno/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/openai-used-frontier-models-to-help-design-its-jalapeno/</guid>
      <pubDate>Wed, 18 Nov 2026 17:45:55 GMT</pubDate>
      <description>OpenAI engineers used frontier models to accelerate development of the Jalapeño inference chip, with model-proposed changes cutting cell area in several blocks.</description>
      <category>inference</category>
      <category>ai-agents</category>
    </item>
    <item>
      <title>Gemini app comes to eligible Android (Go edition) devices</title>
      <link>https://blog.oxelan.com/ai/gemini-app-comes-to-eligible-android-go-edition-devices/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/gemini-app-comes-to-eligible-android-go-edition-devices/</guid>
      <pubDate>Wed, 18 Nov 2026 17:44:28 GMT</pubDate>
      <description>Google is bringing the Gemini app to eligible Android (Go edition) devices and re-architected it for hardware with less memory.</description>
      <category>inference</category>
      <category>ai-agents</category>
    </item>
    <item>
      <title>OpenAI introduces GPT-6.2 Astra and Sol, cutting Astra token prices</title>
      <link>https://blog.oxelan.com/ai/openai-introduces-gpt-6-2-astra-and-sol-cutting-astra-token/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/openai-introduces-gpt-6-2-astra-and-sol-cutting-astra-token/</guid>
      <pubDate>Wed, 18 Nov 2026 02:38:23 GMT</pubDate>
      <description>OpenAI introduced GPT-6.2 Astra, a frontier model, and GPT-6.2 Sol, an efficient model for everyday work; Astra&apos;s input and output token prices are 20% lower than GPT-6 Astra&apos;s.</description>
      <category>inference</category>
      <category>ai-agents</category>
      <category>ai-governance</category>
    </item>
    <item>
      <title>Gemini app adds 66 regional voices on Android and iOS</title>
      <link>https://blog.oxelan.com/ai/gemini-app-adds-66-regional-voices-on-android-and-ios/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/gemini-app-adds-66-regional-voices-on-android-and-ios/</guid>
      <pubDate>Mon, 16 Nov 2026 19:42:32 GMT</pubDate>
      <description>Google is rolling out 66 new regional voices in the Gemini app on Android and iOS, including several English variants plus Hindi, Brazilian Portuguese, Indonesian, and Korean.</description>
      <category>inference</category>
      <category>ai-agents</category>
    </item>
    <item>
      <title>OpenAI expands free ChatGPT access to academic researchers</title>
      <link>https://blog.oxelan.com/ai/openai-expands-free-chatgpt-access-to-academic-researchers/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/openai-expands-free-chatgpt-access-to-academic-researchers/</guid>
      <pubDate>Mon, 16 Nov 2026 17:45:03 GMT</pubDate>
      <description>OpenAI is expanding its academic researcher program to give up to 200,000 researchers free access to ChatGPT, with 50,000 new grants opening November 23.</description>
      <category>ai-governance</category>
      <category>inference</category>
    </item>
    <item>
      <title>SANSAR cryptominer falsifies GPU telemetry on compromised Ray hosts</title>
      <link>https://blog.oxelan.com/security/sansar-cryptominer-falsifies-gpu-telemetry-on-compromised/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/security/sansar-cryptominer-falsifies-gpu-telemetry-on-compromised/</guid>
      <pubDate>Mon, 16 Nov 2026 11:39:39 GMT</pubDate>
      <description>Elastic Security Labs is tracking a Linux cryptomining intrusion that reached root through an exposed Ray job agent and used a falsified NVIDIA library to hide GPU work.</description>
      <category>malware</category>
      <category>incident-response</category>
      <category>inference</category>
    </item>
    <item>
      <title>DeepSeek API adds app features: web search, file input and safety checks</title>
      <link>https://blog.oxelan.com/ai/deepseek-api-adds-app-features-web-search-file-input-and/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/deepseek-api-adds-app-features-web-search-file-input-and/</guid>
      <pubDate>Fri, 13 Nov 2026 15:47:53 GMT</pubDate>
      <description>DeepSeek has exposed the capabilities behind its app through the API, with V4.1-Flash supporting web search, file input, URL reading and safety checks.</description>
      <category>inference</category>
      <category>llm-security</category>
    </item>
    <item>
      <title>Google and Adobe expand partnership to embed tools and AI models</title>
      <link>https://blog.oxelan.com/ai/google-and-adobe-expand-partnership-to-embed-tools-and-ai/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/google-and-adobe-expand-partnership-to-embed-tools-and-ai/</guid>
      <pubDate>Fri, 13 Nov 2026 07:42:16 GMT</pubDate>
      <description>Google said at Adobe MAX that it is expanding its partnership with Adobe, bringing Adobe tools into Google products and Google models into Adobe apps.</description>
      <category>ai-agents</category>
      <category>inference</category>
    </item>
    <item>
      <title>Google&apos;s Gemini Decisions API targets fast closed-set decisions</title>
      <link>https://blog.oxelan.com/ai/google-s-gemini-decisions-api-targets-fast-closed-set/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/google-s-gemini-decisions-api-targets-fast-closed-set/</guid>
      <pubDate>Fri, 13 Nov 2026 03:43:16 GMT</pubDate>
      <description>Google introduced the Gemini Decisions API and the Gemini Decision 1 model for closed-set decisions, returning a confidence score with every answer.</description>
      <category>inference</category>
      <category>gateway</category>
    </item>
    <item>
      <title>Gemini 4 Hydrogen opens to developers with four thinking levels</title>
      <link>https://blog.oxelan.com/ai/gemini-4-hydrogen-opens-to-developers-with-four-thinking/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/gemini-4-hydrogen-opens-to-developers-with-four-thinking/</guid>
      <pubDate>Fri, 13 Nov 2026 01:51:19 GMT</pubDate>
      <description>Gemini 4 Hydrogen is available to developers starting today, with four configurable thinking levels and the option to run alongside Gemini 4 Carbon and Gemini 4 Argon.</description>
      <category>ai-agents</category>
      <category>inference</category>
    </item>
    <item>
      <title>Microsoft 365 outage hits Outlook, Teams and Copilot</title>
      <link>https://blog.oxelan.com/security/microsoft-365-outage-hits-outlook-teams-and-copilot/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/security/microsoft-365-outage-hits-outlook-teams-and-copilot/</guid>
      <pubDate>Thu, 12 Nov 2026 20:41:09 GMT</pubDate>
      <description>A worldwide Microsoft 365 outage affected Outlook, Teams and Copilot, and Microsoft traced the elevated load to servers becoming unexpectedly unavailable rather than to a traffic increase.</description>
      <category>incident-response</category>
      <category>inference</category>
    </item>
    <item>
      <title>Google introduces Gemini 4 Hydrogen with a 1 million-token context window</title>
      <link>https://blog.oxelan.com/ai/google-introduces-gemini-4-hydrogen-with-a-1-million-token/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/google-introduces-gemini-4-hydrogen-with-a-1-million-token/</guid>
      <pubDate>Thu, 12 Nov 2026 18:44:51 GMT</pubDate>
      <description>Google introduced Gemini 4 Hydrogen on November 12, 2026, with a 1 million-token context window and pricing of $1.50 per million input tokens and $7.50 per million output tokens.</description>
      <category>ai-agents</category>
      <category>inference</category>
    </item>
    <item>
      <title>Google Search rolls out Gemini 4 Hydrogen worldwide</title>
      <link>https://blog.oxelan.com/ai/google-search-rolls-out-gemini-4-hydrogen-worldwide/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/google-search-rolls-out-gemini-4-hydrogen-worldwide/</guid>
      <pubDate>Thu, 12 Nov 2026 18:42:58 GMT</pubDate>
      <description>Google has made Gemini 4 Hydrogen available globally in Google Search, describing it as the biggest model upgrade to Search yet.</description>
      <category>inference</category>
      <category>ai-agents</category>
    </item>
    <item>
      <title>Notion customer data exposed in Fireworks AI breach</title>
      <link>https://blog.oxelan.com/security/notion-customer-data-exposed-in-fireworks-ai-breach/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/security/notion-customer-data-exposed-in-fireworks-ai-breach/</guid>
      <pubDate>Thu, 12 Nov 2026 16:42:37 GMT</pubDate>
      <description>According to SecurityWeek, a cyberattack at Fireworks AI exposed some customer data, and Notion is notifying a small set of customers whose Notion AI chat transcripts were involved.</description>
      <category>data-breach</category>
      <category>incident-response</category>
      <category>inference</category>
    </item>
    <item>
      <title>SGLang RCE breach at NEAR AI reaches 18 agents and about 100 users</title>
      <link>https://blog.oxelan.com/security/sglang-rce-breach-at-near-ai-reaches-18-agents-and-about/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/security/sglang-rce-breach-at-near-ai-reaches-18-agents-and-about/</guid>
      <pubDate>Tue, 10 Nov 2026 19:45:51 GMT</pubDate>
      <description>NEAR AI says an attacker exploited a remote code execution flaw in SGLang, reached 18 AI agents, and downloaded account details for about 100 users.</description>
      <category>inference</category>
      <category>llm-security</category>
      <category>incident-response</category>
    </item>
    <item>
      <title>MiniMax releases H3-Image, an open-weight 4K image model</title>
      <link>https://blog.oxelan.com/ai/minimax-releases-h3-image-an-open-weight-4k-image-model/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/minimax-releases-h3-image-an-open-weight-4k-image-model/</guid>
      <pubDate>Tue, 10 Nov 2026 18:40:25 GMT</pubDate>
      <description>MiniMax released H3-Image, an open-weight image generation model that produces a full-resolution 4K image in under 10 seconds, with weights released and access through the MiniMax API Platform and Hailuo AI.</description>
      <category>open-weights</category>
      <category>inference</category>
    </item>
    <item>
      <title>Gemini 4 Carbon ships for developers and Ultra subscribers</title>
      <link>https://blog.oxelan.com/ai/gemini-4-carbon-ships-for-developers-and-ultra-subscribers/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/gemini-4-carbon-ships-for-developers-and-ultra-subscribers/</guid>
      <pubDate>Tue, 10 Nov 2026 16:56:38 GMT</pubDate>
      <description>Gemini 4 Carbon was made available on November 10, 2026, with API access for paying developers and enterprises and a rollout to Google AI Ultra subscribers.</description>
      <category>inference</category>
      <category>llm-security</category>
    </item>
    <item>
      <title>Google DeepMind launches MKA, a benchmark for visually complex knowledge work</title>
      <link>https://blog.oxelan.com/ai/google-deepmind-launches-mka-a-benchmark-for-visually/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/google-deepmind-launches-mka-a-benchmark-for-visually/</guid>
      <pubDate>Tue, 10 Nov 2026 16:54:25 GMT</pubDate>
      <description>Google DeepMind and Artificial Analysis released MKA, a benchmark of 200 tasks that scores agents on visually complex professional work.</description>
      <category>ai-agents</category>
      <category>inference</category>
    </item>
    <item>
      <title>Claude Mythos 5.2 improves CVE Rediscovery and sandbox containment</title>
      <link>https://blog.oxelan.com/ai/claude-mythos-5-2-improves-cve-rediscovery-and-sandbox/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/claude-mythos-5-2-improves-cve-rediscovery-and-sandbox/</guid>
      <pubDate>Tue, 10 Nov 2026 12:40:27 GMT</pubDate>
      <description>According to SecurityWeek, Mythos 5.2 scored 89.6% on CVE Rediscovery and relied on outside credentials or files in 3.1% of sandbox scenarios, while Fable 5.2 carries misuse protections and routes risky cyber tasks to Opus 4.8.</description>
      <category>llm-security</category>
      <category>inference</category>
    </item>
    <item>
      <title>Anthropic ships Claude Fable 5.2 and Claude Mythos 5.2 with cyber benchmark scores</title>
      <link>https://blog.oxelan.com/ai/anthropic-ships-claude-fable-5-2-and-claude-mythos-5-2-with/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/anthropic-ships-claude-fable-5-2-and-claude-mythos-5-2-with/</guid>
      <pubDate>Tue, 10 Nov 2026 08:39:37 GMT</pubDate>
      <description>Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 with output up to 30% faster than the previous generation and a 76.3% score on Irregular&apos;s CyScenarioBench.</description>
      <category>llm-security</category>
      <category>ai-agents</category>
      <category>inference</category>
    </item>
    <item>
      <title>Gemini can now generate and edit files inside the chat</title>
      <link>https://blog.oxelan.com/ai/gemini-can-now-generate-and-edit-files-inside-the-chat/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/gemini-can-now-generate-and-edit-files-inside-the-chat/</guid>
      <pubDate>Sun, 08 Nov 2026 14:17:26 GMT</pubDate>
      <description>Google&apos;s Gemini app can now generate, preview, and edit PDFs and other files inside the chat, with PDF editing on web and mobile and other formats on the web.</description>
      <category>inference</category>
      <category>ai-governance</category>
    </item>
    <item>
      <title>Gemini 4 Carbon takes first on an intelligence index, with access starting among cyber defenders</title>
      <link>https://blog.oxelan.com/ai/gemini-4-carbon-takes-first-on-an-intelligence-index-with/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/gemini-4-carbon-takes-first-on-an-intelligence-index-with/</guid>
      <pubDate>Sun, 08 Nov 2026 14:09:22 GMT</pubDate>
      <description>Google&apos;s Gemini 4 Carbon ranks first on v5.0 of the Artificial Analysis Intelligence Index at 59 and reports 88.2% on SEC-bench Pro, with access starting among cyber defenders.</description>
      <category>llm-security</category>
      <category>ai-governance</category>
      <category>inference</category>
    </item>
    <item>
      <title>Gemini Omni 1.2 Flash now GA with 15-second video and lower price</title>
      <link>https://blog.oxelan.com/ai/gemini-omni-1-2-flash-now-ga-with-15-second-video-and-lower/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/gemini-omni-1-2-flash-now-ga-with-15-second-video-and-lower/</guid>
      <pubDate>Thu, 05 Nov 2026 18:40:10 GMT</pubDate>
      <description>Gemini Omni 1.2 Flash is generally available on Google Cloud, generating and editing up to 15 seconds of video at $0.07 per second of 720p video.</description>
      <category>ai-agents</category>
      <category>inference</category>
    </item>
    <item>
      <title>Open-weight models: the bill arrives after the benchmark</title>
      <link>https://blog.oxelan.com/ai/open-weight-models-economics/</link>
      <guid isPermaLink="true">https://blog.oxelan.com/ai/open-weight-models-economics/</guid>
      <pubDate>Fri, 23 Oct 2026 00:00:00 GMT</pubDate>
      <description>Open weights remove a license fee, not the operating cost. The decision hinges on utilization, latency, privacy and the team you can staff.</description>
      <category>open-weights</category>
      <category>inference</category>
    </item>
  </channel>
</rss>
