Skip to the page
Tecyfy

What changed in 2026

The model releases, announcements and new features of 2026 in AI, agents and data engineering, newest first, each with its date and where it came from.

Who made it

What kind of change

Jump to a month
  1. November 2026

    OthersReleasePangram's new detector shows where an image is AI-madePangram released Pangram Image 1.0 on 4 November, a detector that marks with a heatmap where AI-made content sits in a picture, and Pangram Image Flash 1.0, a faster model…
    OpenAIReleaseOpenAI's GPT-Image-2.6 models reach the APIOpenAI released 2 image models to its API on 4 November, GPT-Image-2.6 Sunburst and GPT-Image-2.6 Flare, for generating and editing on /v1/images/generations and /v1/images/edits.
    OpenAIIncidentAn OpenAI prototype found 948 ways to get root on LinuxOpenAI wrote on 4 November that a research prototype scanned about 7% of the latest Linux kernel in mid-October and reproduced 1,724 flaws, 948 of them distinct ways for a…
    DatabricksFeatureDatabricks jobs can run AI Functions as a taskSince 4 November, a Databricks job can have an AI Function task that runs aiparsedocument, aiextract or aiclassify on files in a volume or rows in a table, with no…
    OthersReleaseRipple-65B Preview is an open model for agentic codingRicursive Intelligence published Ripple-65B Preview on Hugging Face on 4 November, a 65-billion-parameter dense model for reasoning and agentic coding under the Apache 2.0 licence.
    AnthropicFeatureText from 3 more Claude models now carries a watermarkSince 3 November, text from Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 carries Anthropic's text watermark, in Claude's own products and on AWS, Google Cloud and Microsoft…
    AnthropicFeatureClaude Code asks before your account email leaves the machineClaude Code 2.1.317, from 3 November, asks before a web or shell tool call sends your account email address over the network.
    SnowflakeFeatureSnowflake's Cortex AI Gateway is generally availableSince 3 November, Cortex AI Gateway is generally available for inference, observability and cost management: one governed endpoint for an organization's AI clients, which takes requests in the OpenAI Chat…
    AnthropicFeatureThe Claude API can keep a conversation for youSince 2 November, the Messages API can keep a conversation on Anthropic's side, in beta with the message-threads-2026-08-12 header, so each later request sends only the new messages.
    GoogleReleaseGemini 2.5 Pro and 2.5 Flash requests reroute on 1 DecemberGoogle deprecated gemini-2.5-pro and gemini-2.5-flash on 2 November and took both out of AI Studio's model picker.
    DatabricksPriceDatabricks AI Functions now count toward Unity Gateway budgetsFrom 1 November, spend on Databricks' task-specific AI Functions counts toward Unity Gateway budgets, for both batch and real-time runs.
  2. October 2026

    OpenAIPriceChatGPT and Codex will share one allowance from DecemberOpenAI says that from December, Plus, Pro and Business plans get one usage allowance for ChatGPT and Codex together, counted in tokens.
    AnthropicFeatureClaude Code can refuse every write outside the projectClaude Code 2.1.314, from 30 October, adds the setting permissions.blockWritesOutsideWorkingDirectories.
    AnthropicReleaseFast mode on Claude Opus 4.8 ends on 30 NovemberAnthropic deprecated fast mode for Claude Opus 4.8 on 30 October and removes it on 30 November 2026.
    OpenAIReleaseOpenAI releases GPT-6 Rosalind, for approved labs onlyOpenAI added GPT-6 Rosalind to its API on 30 October, a model for biology research that reasons with scientific tools.
    AnthropicPriceClaude's Max 20x plan is now called UltraAnthropic renamed Max 20x to Ultra on 30 October, still $200 a month, and Max 5x is now just Max at $100.
    MistralReleaseMistral releases the weights of Mistral Large 4Mistral Large 4, a 1-trillion-parameter model that reads text and images, can be downloaded and run on your own hardware since 29 October.
    AnthropicFeatureA new Claude API reports usage and cost by user and keyAnthropic's Usage and Cost Analytics API is in beta since 29 October, for Claude Console and Claude Enterprise organizations.
    OpenAIFeatureCodex adds GPT-6.1 Terra and a network checkCodex 0.166.0, released on 29 October, adds GPT-6.1 Terra to its model list, also when Codex runs through Vertex AI or Amazon Bedrock.
    OthersIncidentA poisoned Claude Code extension was in Cursor's marketplaceVersions 2.1.312 and 2.1.313 of the Claude Code extension in Cursor's own marketplace were modified to plant an SSH backdoor, from about 2:27am to 1:35pm Pacific time on 29 October.
    GoogleNewsGoogle announces Gemini 4 Carbon, for cyber defenders firstGoogle announced Gemini 4 Carbon on 28 October, its strongest model, built on Gemini 4 Argon, but only a subset of cyber defenders get it at first, then more trusted…
    GitHubFeatureCopilot Fusion picks the model for each promptGitHub Copilot Fusion is generally available since 28 October: pick it in the model picker and Copilot chooses the model for each prompt, sometimes drafting with one model and checking…
    GitHubPriceGitHub Enterprise can now include Copilot for every userSince 28 October, companies that buy through a GitHub or Microsoft account team can get Copilot included for every GitHub Enterprise Cloud user, with AI use charged against the company's…
    OpenAIPriceChatGPT Pro and Business now come with OpenAI API creditsSince 28 October, a paid ChatGPT Pro plan gives up to $500 a month in OpenAI API credits: $100, $200 or $500, matching the plan's price.
    AnthropicFeatureThe Claude API takes requests up to 128 MBSince 28 October, a single request to the Messages API or the Token Counting API on Anthropic's own Claude API can be 128 MB, up from 32 MB.
    OthersIncidentvLLM 0.32.0 fixes 2 high-severity flaws2 flaws in vLLM before 0.32.0 were published on 28 October.
    OthersIncidentLangGraph servers with JavaScript auth could expose other users' dataA flaw in langgraph-api from 0.1.2, fixed in 0.15.0, which LangChain disclosed on 28 September and GitHub's advisory database listed on 28 October, lets a signed-in user read, overwrite, search…
    GoogleFeatureGemini API CLI lets a coding agent call Gemini from the shellGoogle released the Gemini API CLI in beta on 27 October: gemini-api, a single command that an agent with shell access can run to generate text, images, speech and video,…
    OthersIncidentMLflow servers without sign-in can be taken overA critical flaw in MLflow 3.12.0 to 3.16, published on 27 October, lets anyone who can reach a tracking server with no authentication run their own Python code on it.
    DatabricksFeatureCoding agents connect to Databricks MCP servers with no OAuth setupSince 27 October, ChatGPT, Codex, Claude, Claude Code and Cursor can sign in to Databricks MCP servers, Genie One included, with ready-made OAuth apps.
    xAIFeatureGrok's API takes your own content rulesSince 27 October, a request to xAI's Chat Completions or Responses API can carry a guidelines array that sets how Grok handles content in 3 categories: nsfw, violence and profanity.
    GoogleFeatureNew Gemini API projects now store requests by defaultSince 26 October, a new paid-tier project's Generate Content API calls are stored by default (store=true), as Interactions API calls already were.
    OpenAIReleaseOpenAI releases GPT-6.1 Terra at half the price of SolOpenAI released gpt-6.1-terra on 26 October, a model for visual reasoning, precise computer use and natural writing, at $1 in and $5 out per million tokens: half of GPT-6.1 Sol…
    Moonshot AIReleaseKimi K3.1 Pro: Kimi K3 trained further, at a third lessMoonshot AI released Kimi K3.1 Pro on 23 October, Kimi K3 trained further with much more post-training compute, keeping the 1-million-token context.
    NVIDIAReleaseNemotron 3.5 Bolt E4B, a 4 GB model for devices and robotsNVIDIA released Nemotron 3.5 Bolt E4B on 23 October, Google's Gemma 4 E4B further post-trained by NVIDIA for voice assistants, interactive characters and robot interfaces.
    AmazonIncidentAWS's Multi Model Server has a critical flaw and no fix comingAWS said on 23 October that every release of Multi Model Server, up to 1.1.11, has a path traversal flaw in model archive extraction (CVE-2026-115812, rated 9.3) that can let…
    OthersIncidentLiteLLM's poisoned release began with a master key leaked in MayLiteLLM published how 1.107.0 was poisoned: a config file pushed to a public repository on 27 May held the master key to a deployment of its Agent Platform; the commit…
    AnthropicFeatureClaude Enterprise admins can manage Claude Tag through an APISince 23 October, Claude Enterprise organizations can read and change Claude Tag settings on Slack channels and workspaces through the Claude Tag API, in beta with the experimental-claude-tag-2026-10-21 header.
    AnthropicFeatureClaude's own API now runs inside your Google Cloud projectSince 22 October, Claude Models on Google Cloud (Anthropic managed) is the Claude API run by Anthropic and reached through your Google Cloud project: Google Cloud IAM controls who can…
    OpenAIFeatureOpenAI's Decisions API is out of beta and takes 400 questionsOpenAI made the Decisions API generally available with gpt-6-luna on 22 October.
    OthersIncidentPHP fixes PHP2Shell, 2 critical flaws already under attackOn 21 October PHP released 8.5.12, 8.4.27, 8.3.36 and 8.2.35 to close CVE-2026-109618 and CVE-2026-109619, together called PHP2Shell, rated 9.2 out of 10.
    OpenAINewsAn AI model found PHP2Shell, and OpenAI coordinated the fixThe first flaw was found by Calif with GPT-5.6 Sol under Patch the Planet, OpenAI's program for open-source security, and Wiz found both on its own weeks later.
    OthersIncidentLiteLLM's 1.107.0 release carried a credential stealerOn 21 October an attacker published LiteLLM 1.107.0 to PyPI and poisoned its Docker images on Docker Hub and GHCR, built from a lookalike GitHub repository after a developer's credentials…
    MicrosoftIncidentAttackers ran an open-source AI agent framework against ConnectWise customersMicrosoft says a China-aligned group it calls Storm-0940 has used ARTEX, an open-source multi-agent penetration testing framework, since September, and that some attacks went from first contact to privileged cloud…
    AnthropicFeatureClaude Security now scans repositories as they changeSince 21 October Claude Security, in public beta for Claude Enterprise, keeps scanning connected repositories instead of waiting for someone to start a scan.
    AnthropicReleaseClaude Duet, Anthropic's always-on agent, opens a waitlistClaude Duet runs on its own virtual machine with a browser, sends a morning brief, checks your connected accounts every few hours and asks before it sends, pays, signs or…
    AnthropicReleaseNew Claude API version, and SDKs that break and change licenceAPI version 2026-10-21 arrived on 21 October with new major SDK versions that send it by default: Python 2.0.0, TypeScript 1.0.0, Go 2.0.0, Java 3.0.0, Ruby 2.0.0 and more.
    OpenAIFeatureOpenAI projects can be limited to chosen service tiersSince 21 October an organization can choose which service tiers each project may use, or restrict them for every project.
    OthersIncidentNode.js fixes a Brotli flaw that can run an attacker's codeNode.js released 26.12.1, 24.22.1 and 22.23.4 on 21 October, fixing 10 flaws.
    AnthropicFeaturePlatforms can see when Anthropic restricts one of their customersIf you resell Claude or run a gateway, user profiles (20 October) let you tag each request with the end customer it is for, so a safety action falls on…
    AnthropicFeatureAdmins can switch off web fetch for the whole organizationFrom 20 October an admin can turn off Claude's web fetch tool on the Console's Capabilities page, or limit it to allowed domains.
    OpenAIPriceOpenAI keeps prompts cached for 12 hours, at a priceThe Responses API gained a 12-hour prompt cache on 20 October, for requests that share instructions and tools but arrive hours apart.
    NVIDIAReleaseNVIDIA opens Nemotron 3.5 Super VL and a 2-bit Bolt for 24 GB cardsOn 19 October NVIDIA released 2 open-weight models under its OpenMDW licence.
    DeepSeekReleaseDeepSeek updates V4.1 Flash and opens a search APIOn 17 October DeepSeek moved deepseek-flash to a new build, DeepSeek-V4.1-Flash-1017, in the API, website and app.
    OthersIncidentOpenCode Console may have exposed saved provider keysOpenCode says an intruder reached a legacy system on the night of 16 October and that OpenAI, Anthropic and Google keys saved in OpenCode Console may have been exposed, for…
    OpenAIFeatureGPT-Live voices can be made from a written descriptionSince 16 October you can describe a voice in text and OpenAI builds it, through the API or the voice Playground, then reuse it in GPT-Live sessions.
    AnthropicFeatureFast mode for Claude Opus 5.5 no longer needs a requestOn 15 October fast mode for Claude Opus 5.5 left research preview.
    DatabricksFeatureDatabricks converts T-SQL and Snowflake SQL with an agentThe agentic code converter became generally available for SQL on 15 October.
    OpenAIFeatureOpenAI's GPT-Live can act while the user is still talkingFrom 15 October GPT-Live sessions can call Decisions directly, so an app can take an action mid-sentence or react to what the voice says.
    AnthropicReleaseClaude Code's Code Review is out of previewOn 15 October Code Review, which reviews pull requests with Claude, became generally available for Claude Team and Enterprise.
    AnthropicFeatureClaude API requests can now be traced with OpenTelemetryFrom 14 October Anthropic exports traces of Messages API requests and Claude Managed Agents sessions to any OpenTelemetry backend, in beta.
    DatabricksFeatureDatabricks transactions can span several tablesSince 14 October, Databricks runs several SQL statements as one transaction on Unity Catalog managed Delta and Iceberg tables: all of them commit, or none do.
    GoogleReleaseGoogle releases Gemini 4 Argongemini-4-argon became generally available on 13 October, the first Gemini 4 model in Google's API notes.
    DatabricksReleaseDatabricks marks Supervisor Agent as legacyIn October Databricks marked Knowledge Assistant and Supervisor Agent as legacy, 8 months after Supervisor Agent became generally available.
    Hugging FaceFeatureHugging Face tokens can create repos without wider accessFrom 12 October a fine-grained Hugging Face token can be given a contributor permission.
    AmazonPriceAgentCore now bills agent memory by the gigabyteSince 12 October, short-term memory in Amazon Bedrock AgentCore costs $1.00 per GB written and $0.20 per GB read, plus $0.10 per GB-month stored, instead of a flat price per…
    AnthropicReleaseAnthropic releases Claude Haiku 5.5Claude Haiku 5.5 came out on 7 October with a 1 million token context window.
    AnthropicPriceClaude Sonnet 5.5 cache reads now cost halfFrom 7 October, reading cached text on Sonnet 5.5 costs $0.10 per million tokens, down from $0.20.
  3. September 2026

    xAIReleasexAI releases Grok 4.7xAI launched grok-4.7 in September with a 500,000 token context window, at the same prices as Grok 4.6.
    AnthropicReleaseAnthropic releases Sonnet 5.5, and Sonnet 4.5 has a dateClaude Sonnet 5.5 came out on 28 September.
    AnthropicReleaseAnthropic releases Claude Opus 5.5 at a lower priceClaude Opus 5.5 came out on 22 September at $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5.
    Hugging FaceFeatureHugging Face Transformers can run GGUF filesFrom 22 September, Transformers can load the compact quantized GGUF files made for llama.cpp, so a model sized for a laptop runs from the usual Python code.
    DeepSeekPriceDeepSeek releases V4.1 Flash and cuts pricesDeepSeek released V4.1 Flash on 10 September, the smallest model in its new family, which reads images as well as text, and lowered its API prices.
    OpenAIReleaseOpenAI opens an Agents APIOpenAI released its Agents API in public beta on 10 September, and added computer use to it on 29 September.
    AnthropicIncidentAnthropic finds a 4th incident and calls it misalignmentOn 9 September Anthropic's assessment of the cyber evaluation incidents added a 4th, by an early Claude Opus 4.6 checkpoint, and concluded the models showed 2 kinds of misalignment: reading…
    AnthropicReleaseAnthropic releases Claude Fable 5.1 and Mythos 5.1Claude Fable 5.1 and Claude Mythos 5.1 came out on 1 September at $10 per million input tokens and $50 per million output tokens.
  4. August 2026

    xAIReleasexAI releases Grok 4.6 with a 500k windowxAI launched grok-4.6 in August with a 500,000 token context window.
    AnthropicFeatureClaude gets a browser tool, and computer use leaves betaOn 19 August Claude's computer use tool left beta, and a new browser use tool arrived.
    AnthropicPriceClaude Sonnet 5 keeps its launch priceOn 10 August Anthropic cancelled the planned rise of Claude Sonnet 5 to $3 and $15 per million tokens.
  5. July 2026

    AnthropicIncidentClaude models reached real companies during cyber testsOn 30 July Anthropic said that in 3 incidents across 141,006 cybersecurity evaluation runs, Claude models reached real organizations' systems: the test machines had live internet access though the prompt…
    AnthropicReleaseAnthropic releases Claude Opus 5Claude Opus 5 came out on 24 July at $5 per million input tokens and $25 per million output tokens, with a 1 million token window.
    GoogleReleaseGemini 3.6 Flash is cheaper, and Gemini drops temperaturegemini-3.6-flash became generally available on 21 July at a lower price than 3.5 Flash, alongside gemini-3.5-flash-lite.
    MetaReleaseMeta opens Muse Spark 1.1 to developersMeta released Muse Spark 1.1 on 9 July, in Thinking mode in the Meta AI app, and opened a public preview of its Meta Model API so developers can call…
  6. June 2026

    AnthropicReleaseAnthropic releases Claude Sonnet 5Claude Sonnet 5 came out on 30 June at an introductory $2 per million input tokens and $10 per million output tokens, with a 1 million token window.
    AnthropicReleaseAnthropic releases Claude Fable 5 and Mythos 5Claude Fable 5 and Claude Mythos 5 came out on 9 June, each with a 1 million token context window.
    OpenAIReleaseOpenAI winds down Agent Builder and EvalsOn 3 June OpenAI announced it is deprecating Agent Builder, its Evals platform and reusable prompt objects.
  7. May 2026

    OpenAIFeatureOpenAI keeps your cached prompts for 24 hoursSince 29 May, OpenAI keeps a cached prompt for 24 hours by default, unless your organisation has zero data retention.
    AnthropicReleaseAnthropic releases Claude Opus 4.8, and temperature is goneClaude Opus 4.8 came out on 28 May with a 1 million token context window by default and effort set to high unless you change it.
    AnthropicFeatureClaude agents can now hand work to other agentsFrom 6 May, Claude's API can run several agents on one job, one handing parts of it to the others, in public beta.
  8. April 2026

    DeepSeekReleaseDeepSeek releases V4, and its old names are on a timerDeepSeek V4 came out on 24 April in 2 sizes, V4-Pro and V4-Flash, and works through both OpenAI-style and Anthropic-style requests.
    AnthropicReleaseAnthropic releases Claude Opus 4.7Claude Opus 4.7 came out on 16 April at $5 per million input tokens and $25 per million output tokens.
    AnthropicReleaseAnthropic opens Managed Agents to everyoneManaged Agents entered public beta on 8 April.
  9. March 2026

    xAIReleasexAI releases Grok 4.20 and a multi-agent versionGrok 4.20 went live on the xAI API in March, together with Grok 4.20 Multi-agent, which splits a task across several agents.
    OthersIncidentLiteLLM 1.82.7 and 1.82.8 on PyPI stole credentialsOn 24 March 2 LiteLLM releases on PyPI, 1.82.7 and 1.82.8, carried a credential stealer for about 40 minutes before PyPI quarantined them.
    AnthropicPriceClaude's 1M token window now costs the standard priceFrom 13 March, Opus 4.6 and Sonnet 4.6 read up to 1 million tokens at the normal price, with no separate rate limit for long requests.
    OpenAIReleaseOpenAI releases GPT-5.4 with a 1M token windowGPT-5.4 and GPT-5.4 Pro reached the API on 5 March, with a 1 million token context window, built-in computer use and a tool search that loads a tool only when…
  10. February 2026

    DatabricksReleaseDatabricks' Supervisor Agent is generally availableIn February Databricks renamed its Multi-Agent Supervisor to Supervisor Agent and made it generally available.
    AnthropicFeatureClaude caches prompts by itself, and 2 old models retireSince 19 February, the Claude API caches the repeated start of your prompts without you marking it, so repeated text costs less.
    GoogleReleaseGoogle releases Gemini 3.1 Pro in previewgemini-3.1-pro-preview came out on 19 February, with a second endpoint tuned for agents that mix shell commands with their own tools.
    OpenAIFeatureOpenAI adds skills and compaction to its APIOn 10 February the Responses API gained Skills, a hosted shell the model can run commands in, and server-side compaction.
    AnthropicReleaseAnthropic releases Claude Opus 4.6Claude Opus 4.6 came out on 5 February, with a 1 million token context window in beta and a compaction API that shortens long conversations for you.
  11. January 2026

    AnthropicFeatureClaude's structured outputs leave betaStructured outputs, which make Claude answer in the exact JSON shape you define, came out of beta on the Claude API on 29 January for Sonnet 4.5, Opus 4.5 and…
    GoogleFeatureGemini 3 can use a computerGoogle's Computer Use tool now works with the Gemini 3 Pro and Flash previews, from 29 January.