AI tooling and models

1. OpenAI's named replacement for gpt-3.5-turbo costs 5.5 times as much on a typical mix, and a cheaper tier costs about half

   19 Oct

   OpenAI's deprecations page maps gpt-3.5-turbo-0125 and ft-gpt-3.5-turbo to gpt-5.6-terra, and gpt-4.1-nano to gpt-5.6-luna, for removal on 23 October. The model pages list gpt-3.5-turbo at $0.50 input and $1.50 output per million tokens, Terra at $2 and $12, GPT-5.6 Luna at $0.20 and $1.20, and gpt-4.1-nano at $0.10 and $0.40. A separate 14 October notice names gpt-6-luna, at $0.10 and $0.50, as the replacement for GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini. On a month of 10 million input tokens and 2 million output tokens, gpt-3.5-turbo costs $8.00, Terra costs $44.00 (5.5 times as much), GPT-5.6 Luna costs $4.40 (0.55 times) and gpt-6-luna costs $2.00 (0.25 times).

   So what

   Treat the named replacement as the default to test, not an instruction. For a high volume, low cost workload the price is most of the reason you chose the old model, so run your evals on Luna and on gpt-6-luna before you accept Terra, and decide on measured quality rather than on the mapping table. The same applies to gpt-4.1-nano: the named landing, GPT-5.6 Luna, is 2.4 times the price on that mix ($1.80 to $4.40), where gpt-6-luna is 1.1 times ($2.00). Two caveats keep the arithmetic honest. These are reasoning models, and OpenAI's reasoning guide says reasoning tokens are billed as output tokens, so measure output tokens per request on real traffic. And set reasoning.effort deliberately: the Terra and Luna pages list none, low, medium (the default), high, xhigh and max.

   Source: OpenAI, GPT-5.6 Terra model pageIssue 7 · Week of 19 October 2026

2. Every o1 and GPT-4 Turbo migration lands on GPT-5.6 Sol, and its $4 and $20 price is promotional through at least 21 November

   19 Oct

   OpenAI's deprecations page names gpt-5.6-sol as the replacement for o1, o1 pro, o3-mini, GPT-4, GPT-4 Turbo, gpt-4-1106-preview and gpt-4o-2024-05-13, all removed on 23 October. The Sol model page lists $4 input, $0.40 cached input and $20 output per million tokens and says "GPT-5.6 Sol's promotional pricing is available at least through November 21, 2026", describing it as a 20% reduction in input pricing and a 33% reduction in output pricing. That implies earlier prices of about $5 and $30. Prompts over 272K input tokens are priced at 2x input and 1.5x output for the whole request. On a month of 10 million input and 2 million output tokens, Sol costs $80.00, against $270.00 for o1, $2,700.00 for o1 pro, $160.00 for GPT-4 Turbo, $420.00 for GPT-4 and $19.80 for o3-mini.

   So what

   The saving is real for o1, o1 pro, GPT-4 and GPT-4 Turbo, and it survives the old price: at $5 and $30 the same month costs $110.00, still 59 percent less than o1, 96 percent less than o1 pro, 74 percent less than GPT-4 and 31 percent less than GPT-4 Turbo. The loss is on o3-mini, where Sol is 4.0 times the cost today and 5.6 times at the earlier price. So do not book the savings until OpenAI says the promotion is permanent, and route o3-mini traffic to Terra first, which is the model OpenAI names for o4-mini and costs 2.2 times o3-mini on this mix. Put 21 November in your calendar either way.

   Source: OpenAI, GPT-5.6 Sol model pageIssue 7 · Week of 19 October 2026

3. Your coding agent's login file is a production secret, and GitHub now treats Anthropic keys that way by default

   18 Oct

   StepSecurity reports that OpenCode stores the provider credentials a developer links at ~/.local/share/opencode/auth.json on macOS and Linux. On 18 October OpenCode said a breach it ties to a large scale distillation attack affected 1.4% of its Console users, those with external providers linked. On 13 October GitHub added Anthropic's two API key types to secret scanning and made push protection block them by default: "Repositories with secret scanning enabled, including free public repositories, will have commits containing these secrets automatically blocked."

   So what

   A coding agent's login file belongs in the same inventory as your production keys, because it holds the same power. List the agents your team runs and the provider keys each one can reach, give each developer their own key per provider rather than a shared team key, and set a spend limit where the provider offers one, so a stolen key shows up as a capped bill instead of an open one. GitHub's block is a good default and it only covers pushes: a key already in your history, in a log or in a chat paste is not caught by it, so scan your history once and rotate anything that was ever committed.

   Source: GitHub, secret scanning blocks Anthropic keys by defaultIssue 7 · Week of 19 October 2026

4. Haiku 5.5 costs a tenth of Haiku 4.5 per token, until a prompt passes 100,000 tokens

   7 Oct

   Anthropic launched Claude Haiku 5.5 on 7 October. For prompts up to 100,000 tokens it costs $0.10 input and $0.50 output per million tokens, against $1 and $5 for Haiku 4.5. For prompts over 100,000 tokens it costs $0.50 and $2.50. The pricing page says the prompt length of a request "counts all of its input tokens, including cache reads and cache writes", that each request is priced on its own, and that a request over the threshold pays the higher prices "even when part of its prompt is a cache hit". The migration guide adds that Haiku 5.5 uses the same tokenizer as Claude 4.7 and later, which produces approximately 30 percent more tokens for the same text than Haiku 4.5.

   So what

   Do the arithmetic on your own traffic before you promise a saving. A tenth of the rate times about 1.3 times the tokens is roughly 0.13 of the old bill, so about 87 percent cheaper, for prompts under the line. Above it the rate is half of Haiku 4.5, the same arithmetic gives about 0.65, and the saving drops to about 35 percent. Look at the long tail of your prompt lengths rather than the average, because agent loops grow: the migration guide says thinking blocks from earlier turns stay in context and count as input, where Haiku 4.5 kept only the latest turn. Compare the shape of the line as well as the rate. OpenAI lists GPT-6 Luna at the same $0.10 and $0.50, but puts its short context limit at 272K input tokens and charges $0.20 and $0.75 beyond it, where Haiku 5.5 steps up fivefold on both. And do not rush the move on a deadline that does not exist yet. The deprecations page lists Haiku 4.5 as active with a tentative retirement of "Not sooner than October 15, 2026", and Anthropic commits to at least 60 days of notice before retiring a public model.

   Source: Anthropic, pricingIssue 6 · Week of 12 October 2026

5. Three labs now list a model at $2 and $10, and the Google one doubles on 1 December

   13 Oct

   Google made Gemini 4 Argon (gemini-4-argon) generally available on 13 October, describing it as its most capable model for complex coding, agentic, math and science tasks. The pricing page lists it at $2 input and $10 output per million tokens "through November 30, 2026" and $4 and $20 "starting December 1, 2026", with output pricing that includes thinking tokens. Context caching goes from $0.10 to $0.20, batch from $1 and $5 to $2 and $10, and Priority from $3.60 and $18 to $7.20 and $36. Only implicit caching is supported, and the model is not available on the free tier or the Google AI Pro plan. That puts it level with Claude Sonnet 5.5 and GPT-6.1 Sol, both listed at $2 and $10 on their own pricing pages today, with no change scheduled on the pages we read.

   So what

   Cost your product at $4 and $20, not at $2 and $10. A launch price that is scheduled to double is an acquisition offer, and anything you build on it, whether a free tier, a fixed price plan or a margin calculation, inherits the December price whether you planned for it or not. If Argon wins your evals, the honest comparison on 1 December is $4 and $20 against Sonnet 5.5 and GPT-6.1 Sol at $2 and $10, so ask whether the quality gap is worth double. Two other details move the bill: the output price includes thinking tokens, and explicit caching is not supported, so cache savings depend on implicit hits you cannot arrange.

   Source: Google, Gemini API pricingIssue 6 · Week of 12 October 2026

6. Cache reads on Sonnet 5.5 are now 5 percent of the input price, so caching comes first

   7 Oct

   On 7 October Anthropic lowered Claude Sonnet 5.5 prompt cache reads from $0.20 to $0.10 per million tokens, which its release notes describe as "0.05x the base input price instead of 0.1x", with cache writes and all other prices unchanged. The pricing page lists Claude Opus 5.5 at the same 0.05x multiplier. Cache writes stay at 1.25x the base input price for the five minute duration and 2x for one hour. That brings the Sonnet 5.5 read to the same $0.10 as the cached input rate on GPT-6.1 Sol, which we reported in issue 5 when we said Anthropic at $0.20 against OpenAI at $0.10 could decide the choice.

   So what

   Correct that comparison before you reuse it: on cached input the two vendors now match, and Gemini 4 Argon lists $0.10 for implicit caching until 30 November and $0.20 after, so the three way comparison is level only until December. More usefully, the arithmetic now favours caching early. On the five minute duration a prefix pays for itself after a single read, because 1.25 plus 0.05 is 1.30 times the base price against 2.00 for two uncached sends. A prefix sent in eleven requests, one write and ten reads, costs 1.75 times one uncached send instead of 11, about 84 percent less on those input tokens. Put the stable material first, meaning the system prompt, tool definitions and reference documents, keep anything that changes per request at the end, and check the usage block in each response to confirm you are getting hits.

   Source: Anthropic, API release notesIssue 6 · Week of 12 October 2026

7. Claude Haiku 5.5 launched with a 1M token window, and a migration that is not a model id swap

   7 Oct

   Anthropic launched Claude Haiku 5.5 (claude-haiku-5-5) on 7 October with a 1M token context window, 128K maximum output and adaptive thinking controlled by the effort parameter. The model id is fixed, with no date suffix and no alias. Compared with Haiku 4.5, a request now returns a 400 error if it sends manual thinking with budget_tokens, sets top_k at all or sets temperature or top_p to anything but their defaults, ends the messages with an assistant prefill even when thinking is off, or declares the computer_20250124 tool. Sending a thinking block back after changing the system prompt, the tools or earlier messages also returns a 400. Thinking text is omitted unless you set thinking.display to "summarized". The guide says Haiku 5.5 runs safety classifiers that can decline a request, that it has no server side fallback, and that Priority Tier is not supported.

   So what

   Treat this as a migration, not a model id swap. Search your code for budget_tokens, temperature, top_p, top_k and any request that ends on an assistant turn, because each one is now a hard failure rather than a quiet change. Select response content blocks by type, not by position, since a response can begin with a thinking block even when you never asked for thinking. Raise max_tokens, because thinking counts toward it and a small limit can stop after a thinking block with no text. Handle the refusal stop reason, and if you rely on Priority Tier for Haiku 4.5, plan that capacity separately. On Amazon Bedrock, structured outputs are not available for this model.

   Source: Anthropic, Haiku 5.5 migration guideIssue 6 · Week of 12 October 2026

8. Managed Agents can now write and run their own multi phase workflows, in beta

   9 Oct

   On 9 October Anthropic put dynamic workflows into beta for Claude Managed Agents. You enable them in the agent definition with multiagent set to {"type": "multiagent_20261001", "workflows": {"type": "enabled"}}, under the managed-agents-2026-04-01 beta header. The documentation describes a workflow as "a program that runs many agents in phases and combines what they return", which the server runs in the background as a workflow run while the agent keeps working or ends its turn. With the multiagent_20261001 type, workflows and subagents are both enabled by default, and so are inline agents that a workflow defines itself. You can list up to 20 predefined agents for a workflow to use. Separately, on 14 October Anthropic made Managed Agents scheduled deployments available on Claude Platform on AWS.

   So what

   The risk is spend you did not decide on. Choosing the multiagent_20261001 type turns workflows on by default, and a workflow can fan out to many agents, so if you only want delegation, set workflows to {"type": "disabled"} explicitly. Set a session budget when you create the session, because the documentation says you cannot add one to an existing session, and runs pause when the budget is reached. Tell the agent in its system prompt when a run is warranted, as the documentation suggests, and keep every listed agent on the same inference geography as the main agent, which the documentation requires.

   Source: Anthropic, multiagent orchestrationIssue 6 · Week of 12 October 2026

9. Claude OpenTelemetry export is in beta, and it keeps prompts out of your traces unless you switch them on

   14 Oct

   On 14 October Anthropic released OpenTelemetry export for the Claude API in beta, covering Messages API requests, tool uses and Managed Agents sessions. The Claude SDK emits spans in Python, TypeScript, C#, Go, Java, PHP and Ruby, and API tracing is set up as a destination in the Claude Console, which needs Admin permission. The default content mode is metadata only, with spans holding identifiers, timing, status and token counts. Full content mode adds prompts and responses, up to 0.75 MB per span by default before the content is truncated and anthropic.content.truncated is set to true. The documentation says span delivery is best effort and that exported spans are not a complete record of API usage.

   So what

   Start in metadata only mode, which is the default, and make switching to full content a decision rather than an accident: prompts and responses will sit in your observability vendor under its own access and retention settings, which the documentation tells you to check. Do not build billing or quota enforcement on spans, because delivery is best effort and the documentation points to the Usage API for that. And if you add up tokens from traces, sum spans from one source only. The SDK and the API both emit an anthropic.messages.create span for the same request, and adding both double counts.

   Source: Anthropic, OpenTelemetry for the Claude APIIssue 6 · Week of 12 October 2026

10. GitHub Actions workflow lockfiles are in public preview, so a moved tag can no longer change what your workflow runs

    14 Oct

    GitHub put Actions workflow dependency lockfiles into public preview on 14 October for GitHub Enterprise Cloud, Enterprise Cloud with data residency, Team and Free plans. The changelog says "Tags and branches can move after you review a workflow, so a later run might execute different code". A lockfile records the exact commits for actions and their transitive dependencies, including reusable workflows, in .github/workflows/actions.lock, and you create and maintain it with the gh actions-lock extension for the GitHub CLI. Once a workflow is onboarded, Actions requires every dependency to be in the lockfile and runs the recorded commit even if a tag or branch moves. A Require lockfile policy is also in public preview for enterprise, organization and repository administrators, and Dependabot regenerates the matching lockfile entry in the same pull request when it updates an action in an onboarded workflow. GitHub is clear that locking "does not determine whether the dependency itself is safe".

    So what

    Onboard your deploy and release workflows first, because they hold the production secrets and a moved tag does the most damage there. Locking pins what you reviewed, it does not review it, so read each action before you lock it. It also does nothing about a malicious install step inside a workflow, which is the tensorlake case in this issue, so it sits alongside ignore-scripts and a careful look at new dependencies rather than replacing them.

    Source: GitHub, changelogIssue 6 · Week of 12 October 2026

11. Two labs priced a day apart at exactly $2 and $10. The Anthropic cut is nearer 13 percent than 33.

    29 Sep

    Anthropic released Claude Sonnet 5.5 on 28 September at $2 input and $10 output per million tokens, with cache reads at $0.20, against Sonnet 4.5 at $3 and $15 with cache reads at $0.30. OpenAI released GPT-6.1 Sol on 29 September at the same $2 and $10, with cached input at $0.10, a fifth of GPT-6 Astra at $10 and $50. The Anthropic migration guide adds a caveat the price table does not: Sonnet 5.5 uses Sonnet 5's tokenizer, and against Sonnet 4.6, Sonnet 4.5 and Haiku 4.5 the same text produces about 30 percent more tokens, depending on the content. A 2000x1500 image costs about 2.5 times as many tokens, because 5.5 uses the high resolution image tier.

    So what

    Do the arithmetic before you promise anyone a saving. A third off the rate, times about 30 percent more tokens for the same text, is roughly 13 percent cheaper in practice, not 33. Images get more expensive, not less. Anyone already on Sonnet 5 sees no token inflation at all, because it is the same tokenizer. Run your real prompts through the token counting endpoint, then raise max_tokens, or long outputs will start truncating at the old limit. And note where the two vendors differ: OpenAI's cached input is $0.10 against Anthropic's $0.20, which decides it if you resend a large fixed prompt.

    Source: Anthropic, Sonnet 5.5 migration guideIssue 5 · Week of 28 September 2026

12. Mistral gave one day of notice. Reddit gave six weeks. Plan for the notice you actually get.

    30 Sep

    On 29 September Mistral's changelog announced that Leanstral 1.5 and OCR 4.0 were deprecated and retired on 30 September 2026, one day later, with Z.ai GLM 5.2 following on 31 October. On 30 September, according to TechCrunch, Reddit said it will retire RSS feeds on 13 November 2026 and close remaining public API access in March 2027, with third-party app and bot developers required to register before 12 January 2027. Reddit is quoted saying "We know RSS has been a beloved part of the open web for a long time". Separately, Reve says that after 27 September 2026 creating new images and videos is no longer available, with asset downloads until 31 October, and Google is reported to be retiring Gemini Gems on 17 November in favour of a skills format.

    So what

    Four withdrawals in one week, with notice periods ranging from one day to six weeks. The useful response is not outrage, it is an inventory. Write down every external thing your product reads or calls that you do not pay for, because a free public API is a courtesy and courtesies get withdrawn. Then rank that list by how long it would take you to replace, and fix the worst one. A hardcoded model id in a document pipeline is the cheapest thing in the world to protect and the most annoying to discover at 2am.

    Source: Mistral, changelogIssue 5 · Week of 28 September 2026

13. Two frontier labs cut prices on the same Tuesday. Only one of them is a drop-in swap.

    22 Sep

    On 22 September OpenAI released GPT-6 Sol at $2 input and $10 output per million tokens, and GPT-6 Luna at $0.10 and $0.50, for prompts up to 272K input tokens. GPT-5.6 Sol and GPT-5.6 Luna remain listed at $4 and $20, and $0.20 and $1.20. The same day Anthropic released Claude Opus 5.5 at $4 and $20, twenty percent below Opus 5 at $5 and $25, with cache hits at $0.20 because Opus 5.5 prices them at 0.05x base input rather than the usual 0.1x. Anthropic's release notes also list three request shapes that now return a 400 error on Opus 5.5.

    So what

    The OpenAI cut costs you an eval run. The Anthropic one has a refactor attached: the thinking field is rejected in both its enabled and disabled forms, tool_choice types any and tool return 400, and computer use needs the computer_toolset_20260801 toolset on the Claude API and Google Cloud. Search your call sites for all three before you point a single request at Opus 5.5. And read what OpenAI does not say: GPT-5.6 Sol is explicitly promotional at least through 21 November 2026, while nothing on either page labels GPT-6 pricing as promotional at all. Budget on the number, not on the direction of travel.

    Source: OpenAI, API changelogIssue 4 · Week of 21 September 2026

14. Anthropic now bills for some refusals that never produced a token

    24 Sep

    On 24 September Anthropic expanded which refusals are billed to include refusals that arrive before any output, when stop_details.category is bio, frontier_llm or reasoning_extraction, which it describes as the categories where it measures low volumes of false positives. Mid-stream refusals were already billed. The newly billed refusals are charged at the rates of the model that ran them. Refusals before output in other categories are still not billed, fallback credit is unchanged, and the change applies on all platforms.

    So what

    Narrow, defensible, and still worth a look at your own bill. If you run red team suites, jailbreak regression tests or safety evals that deliberately trip those three classifiers, that suite just acquired a line item. Log stop_details.category on every response now, so the charge is attributable when it lands instead of being an unexplained rise in next month's invoice.

    Source: Anthropic, Claude API release notesIssue 4 · Week of 21 September 2026

15. OpenAI shipped a managed Agents API 15 days after shutting down the Assistants API. Keep your own state.

    10 Sep

    On 10 September OpenAI released the Agents API in public beta: a managed harness in which OpenAI runs session orchestration, context compaction and recovery, and sessions persist across turns. OpenAI's documentation says it currently offers data residency only in the US and does not support Zero Data Retention. The Assistants API, OpenAI's previous managed layer for agent state, shut down on 26 August.

    So what

    A managed harness is a real shortcut, and this is the second time OpenAI has offered to hold your agent's state. The first one was retired fifteen days before this one launched. Use it if it saves you time, but write every session item to your own database as it streams and keep agent definitions in your repo. Do not promise customers Zero Data Retention or non-US residency on top of it until the documentation says otherwise.

    Source: OpenAI, Agents API documentationIssue 3 · Week of 14 September 2026

16. OpenAI now lets you expire API keys. Existing keys are unaffected, so the setting alone changes nothing.

    15 Sep

    On 10 September OpenAI added expiration dates for project API keys and let administrators enforce a maximum key lifetime at organisation or project level. On 15 September it added controls to allow only service-account keys, allow only user-owned project keys, or disable new key creation. OpenAI notes that existing keys are unaffected.

    So what

    That last sentence is the whole point. A non-expiring key pasted into an environment variable two years ago stays exactly as dangerous after you flip the new switch. Set a maximum lifetime at organisation level, move production to service-account keys, turn off ad hoc personal key creation, and then actually rotate, because only rotation brings the old keys under the new policy.

    Source: OpenAI, API changelogIssue 3 · Week of 14 September 2026

17. The open-weight commons now has a landlord, and he sells GPUs

    3 Sep

    Nvidia confirmed it will buy Hugging Face for $12.93 billion. Jensen Huang said the platform stays open, keeps supporting open-weight models, and that Nvidia compute will not be required to build or deploy through it. Hugging Face rejected a reported $500M approach from Nvidia last year.

    So what

    Take the commitments at face value and the structural point still stands: the default distribution channel for open models is now owned by the company that sells the hardware they run on. Nothing breaks this week. What changes is that "we can always self-host an open model" is now a dependency on one vendor's continued goodwill rather than on a neutral commons. Worth knowing which of your fallbacks actually resolve to huggingface.co.

    Source: TechCrunchIssue 2 · Week of 7 September 2026

18. A deprecation page is a roadmap, and almost nobody reads it

    OpenAI's deprecations page currently lists seven separate shutdown dates falling between 23 October 2026 and 26 February 2027, covering reasoning models, image models, the audio and realtime family, Whisper, the Evals platform, reusable prompts and Agent Builder.

    So what

    Every one of these was announced months ahead, in public, with a date. The failure mode is never the notice, it is that nobody owns checking. Put the page in a calendar reminder the way you would a certificate expiry, because that is functionally what it is.

    Source: OpenAI, deprecationsIssue 2 · Week of 7 September 2026

19. A $13.3B prompt-to-app company is a market structure story, not a funding story

    12 Aug

    Lovable confirmed a $400M Series C led by Menlo Ventures and the Scaleup Europe Fund at a $13.3B valuation.

    So what

    The interesting number is not the valuation, it is what it buys: a long runway to keep the generated-app defaults improving. If you build on one of these platforms, your production posture is now partly a function of someone else's roadmap. Worth knowing which of your safety properties you own and which you are renting.

    Source: TechCrunchIssue 1 · Week of 31 August 2026

20. The AI funding boom is a US story that gets reported as a global one

    Crunchbase's analysis of the current cycle makes the point directly: the AI startup funding boom is not a global phenomenon, and North American totals are doing most of the work in the headline numbers.

    So what

    If you are raising outside North America, benchmark against your own region rather than the global headline, or you will walk into a room expecting a market that is not there. The averages are being set by a geography you may not be in.

    Source: Crunchbase NewsIssue 1 · Week of 31 August 2026

21. Coding on a phone stopped being a joke

    29 Jun

    Cursor shipped a mobile app for supervising coding agents. Anthropic's head of Claude Code, Boris Cherny, said he had almost entirely switched, describing most of his coding as now happening on his phone.

    So what

    The signal is not the form factor, it is the shift from writing to reviewing. If the job is becoming supervision, your bottleneck moves to how fast someone can tell a good diff from a bad one, which is a code-review and test-coverage problem long before it is a tooling problem.

    Source: TechCrunchIssue 1 · Week of 31 August 2026

22. Coding agents are moving from prompted to always-on

    Cursor introduced Automations for always-on agents, describing hundreds of automations per hour and incident-response use where a PagerDuty alert starts an agent that queries server logs over MCP.

    So what

    An agent with production log access is a new identity in your system. Before you wire one up, answer the boring question: what can it read, what can it change, and where is that written down.

    Source: TechCrunchIssue 1 · Week of 31 August 2026

23. Claude Opus 5 shipped

    Anthropic released Claude Opus 5, reporting more than double Opus 4.8's performance on Frontier-Bench v0.1 at a lower cost per task, and within 0.5% of the top CursorBench 3.2 score at roughly half the cost per task.

    So what

    Cost per task is the number to compare, not cost per token. A model that costs more per token and finishes in fewer attempts can be cheaper in production, which is exactly the comparison a per-token price list hides.

    Source: AnthropicIssue 1 · Week of 31 August 2026

24. Anthropic published a 2026 agentic coding trends report

    A primary-source data set on how coding agents are changing development workflows, rather than another survey write-up of a survey.

    So what

    Worth an hour if you are setting an AI tooling policy this quarter. Read the methodology before the conclusions, as you would with any vendor-published research on its own category.

    Source: Anthropic (PDF)Issue 1 · Week of 31 August 2026

Other topics