5 verified AI, developer and Web3 updates worth testing this week (2026)

Since the last issue, the strongest tryable leads fall into two groups: four software changes that viewers can reproduce with a consumer account, an API key, or a Copilot subscription, and one live trading promotion with public rules but real market exposure. The list is ordered by how easily a creator can turn the update into a repeatable video test. The software updates were published or updated from August 13 to August 14, 2026; the QUID promotion runs through August 21. 1 2 3 4

Quick scan

Pick	What changed	Best fit	Main friction	Safe first test
ChatGPT app updates	Interactive quizzes, project-memory switching, and Think on Free and Go arrived in the August 14 release note. 1	Broad consumer and education audience	Memory settings and platform availability differ by feature	Run the same ten-question quiz, then test a harmless project preference in a new chat
Gemini 3.7 Flash	Google marked the model generally available on August 13, with multimodal input and an introductory price through December 31. 2	Builders testing coding and agent workflows	API billing, quotas, and computer use remain separate constraints	Give it the same synthetic bug report, screenshot, and PDF as a comparison model
Copilot CLI workflow controls	The weekly release added `/tasks`, queued prompts, headless plan-plus-autopilot mode, and `/rewind`. 3	Developers who work in a terminal	Each interaction draws from the plan's AI Credits; agents can still modify files	Use a disposable repository and show the task queue, approval prompts, and rollback
Copilot Memory and Ollama in JetBrains	JetBrains gained cross-session Copilot memory and Ollama as a bring-your-own-key provider. 5	JetBrains developers comparing hosted and local models	Memory needs deliberate data hygiene; local model performance depends on the setup	Store one harmless project preference, reopen a session, then compare it with an Ollama run
Kraken QUID Trading Challenge	A ranked spot-trading promotion offers a share of 160,000 QUID from August 7 through August 21. 4	Existing eligible Kraken Pro users	Minimum volume, KYC, regional exclusions, fees, and price risk	Film the eligibility and rules flow; treat any real trade as a risk test, not a reward claim

1. ChatGPT app updates: a broad audience test

What changed. OpenAI's August 14 release note adds interactive quizzes on web and mobile for all consumer ChatGPT plans and Edu plans. A user can ask ChatGPT to quiz them and answer inside the same conversation. The update also lets eligible users switch an unshared project's memory between default memory and project-only memory. Shared projects remain project-only. 1

The same release note adds the Think option on the web for Free and Go users, plus several Android composer improvements. ChatGPT and Codex are also available in a Linux public preview on selected distributions, with browser actions available while control of other desktop apps remains unsupported. 1

How it works. The quiz feature is a conversation mode rather than a separate course or app. Project memory is a setting attached to an eligible unshared project, so a creator can compare what a fresh conversation remembers under default memory and project-only memory. Think is a web model-selection option for Free and Go accounts. 1

Why viewers may care. This is the easiest item in the list to demonstrate to a general audience. A video can show whether an AI tutor gives useful questions, corrects an answer fairly, and changes difficulty when the user asks. The memory switch also creates a concrete privacy and continuity test instead of another open-ended chat demo.

Who it suits. It suits education, study, productivity, and general-purpose AI channels. It gives a creator a much wider test audience than a feature limited to a developer IDE.

Access and limits. Interactive quizzes cover consumer and Edu plans, but project-memory switching is limited to eligible unshared projects. Shared projects keep project-only memory. The Linux application is a public preview, so it should be presented as a test surface rather than a finished desktop replacement. 1

Safe video test. Prepare ten questions from a public topic and ask ChatGPT to quiz the host one question at a time. Record wrong-answer handling, source-checking, and difficulty changes. In a separate throwaway project, save one harmless preference, switch the memory setting, open a new chat, and ask what context carries over. Keep personal, financial, and client information out of the test.

Where to verify. Use OpenAI's August 14 release note immediately before filming. The page is a live changelog, so the plan and platform labels may change after this issue.

2. Gemini 3.7 Flash: multimodal input with a real API test path

What changed. Google marked Gemini 3.7 Flash generally available on August 13, using the stable model ID gemini-3.7-flash. Google positions it for coding, web development, and agentic workflows, with an introductory price through December 31, 2026. 2

How it works. The model page lists text, image, video, audio, and PDF inputs with text output. It supports a 1,048,576-token input limit and a 65,536-token output limit. It also lists code execution, file search, function calling, search grounding, structured outputs, and computer use as supported capabilities; computer use is still marked Preview. 6

A viewer can open the model in Google AI Studio or use the Gemini API after obtaining an API key. The model page links to both paths and records its latest update as August 13, 2026. 6

Why viewers may care. The filmable difference is the input bundle. A creator can hand the model a screenshot, a short bug-report PDF, and a code file, then show whether the model connects the visual symptom to a workable fix. That test says more than a single text prompt or a leaderboard screenshot.

Who it suits. It suits developers, technical educators, and creators who can show a real document-to-code or agent workflow. It is less useful for a general audience if the video only reports a model score.

Access, cost, and limits. Google describes the launch price as introductory through December 31 rather than permanent. The model page says low, medium, and high thinking are supported while minimal is unsupported, and computer use remains Preview. Check current quotas and billing before recording because an API test can consume credits even when the prompt looks small. 2 6

Safe video test. Create a small synthetic web page with one deliberate layout bug. Provide the same code, screenshot, and PDF bug report to Gemini 3.7 Flash and one comparison model. Track the first diagnosis, the patch it proposes, the number of manual corrections, and the result at two viewport sizes. Do not use private repositories or real credentials.

Where to verify. Start with the Gemini API release notes, then open the Gemini 3.7 Flash model page and the linked Google AI Studio entry point.

3. Copilot CLI: queue work while the agent is still running

What changed. GitHub's weekly release page, published August 14 for the week of August 10, adds several terminal workflow controls. A creator can view subagents and their tasks in /tasks, queue prompts and shell commands while an agent turn is running, combine --plan with --mode autopilot in headless mode, and use /rewind to restore Copilot changes without requiring Git or discarding user edits. 3

How it works. Copilot CLI runs in a terminal and can work with issues, pull requests, branches, and files. The current product page includes /plan, /fleet for parallel subagents, /resume for persistent work, and /IDE to move a session into an editor. The August update is worth separating from those older entry points because the queue and rewind controls are the parts that change the recording workflow. 3 7

Why viewers may care. A terminal agent becomes easier to judge when the video shows its work queue and recovery path. Viewers can see whether parallel tasks save time, whether queued instructions arrive in the right order, and whether a rollback removes the intended change.

Who it suits. It suits developers, coding educators, and technical creators who already use a terminal-based workflow. The CLI supports macOS, Linux, and Windows. 7

Access, cost, and limits. GitHub says Copilot CLI is included in Free, Pro, Pro+, Max, Business, and Enterprise plans. Each interaction draws from the plan's AI Credits allowance. Business and Enterprise administrators must enable the CLI, and the installation path includes npm install -g @github/copilot followed by GitHub authentication. Every file change and command execution still requires explicit approval, so autopilot does not remove the need to review actions. 7

Safe video test. Create a disposable repository with two small tasks. Start one task in /plan mode, queue a second prompt while the first runs, inspect /tasks, and use /rewind after an intentionally visible change. Show the approval prompt and final diff. Keep MCP servers, production repositories, and secrets out of the first recording.

Where to verify. Use the August 10 weekly changelog for the change list and the Copilot CLI page for current plan and installation details.

4. Copilot Memory and Ollama in JetBrains: compare continuity with local models

What changed. GitHub's August 11 JetBrains update adds Copilot memory across agent chat sessions and Ollama as a bring-your-own-key provider. The same release also lets JetBrains users install Copilot CLI automatically from an integrated terminal on macOS, Linux, and Windows. 5

How it works. Copilot memory can be managed with a Copilot Memory toggle in the settings portal. Ollama appears as a provider that can be configured and selected across the JetBrains experience. The release page points viewers to the latest GitHub Copilot plugin for JetBrains. 5

Why viewers may care. These are two different experiments under one familiar IDE. Memory tests whether repeated context actually reduces setup work. Ollama tests how a local model changes response time, capability, and dependence on a hosted provider. A video can keep the prompt and project fixed while changing one variable at a time.

Who it suits. It suits JetBrains developers, local-AI enthusiasts, and technical educators who can explain the difference between a hosted Copilot model and an Ollama-backed model. It is a specialist test rather than a broad consumer feature.

Access and limits. The update requires the current GitHub Copilot plugin for JetBrains and a Copilot account. Ollama adds a local provider, but the result depends on the model and machine used for the test. Memory also changes the data a user chooses to carry between sessions, so a creator should document the setting and keep the test context harmless. 5

Safe video test. Use a throwaway project with a harmless stack preference and one small coding task. Save the preference, start a new agent chat, and record whether Copilot recalls it. Then run the same task through an Ollama model, recording latency, output quality, manual edits, and whether the local setup can complete the task. Do not put client code, access tokens, or personal data into the memory test.

Where to verify. Read the JetBrains update, the linked Copilot memory documentation, and the latest JetBrains plugin versions.

5. Kraken QUID Trading Challenge: a live promotion with a real trading threshold

What changed. Kraken's official rules describe the QUID Trading Challenge 2026 as a ranked promotion, not a random drawing. In Lagos time, the promotion runs from 12:00 on August 7 through 00:59 on August 22, and eligible participants compete for a share of 160,000 QUID by executing QUID spot trades on Kraken Pro. 4

How it works. A participant enrolls through the Promotions area in the Kraken Pro app or web interface. Only QUID spot trades completed after enrollment count. The minimum qualifying volume is 1,500 QUID, and the reward depends on rank: ranks 1 through 200 receive fixed tier amounts, while rank 201 and above share 40,000 QUID in proportion to their qualifying volume. 4

Why viewers may care. This is a verifiable Web3 campaign with a clear end date and a public rule set. The video opportunity is the eligibility and economics walkthrough: what must be traded, when the volume starts counting, how the ranks work, and what a viewer still has to verify in their own account.

Who it suits. It suits existing Kraken Pro users in an eligible jurisdiction who already understand spot trading. It is a poor fit for a viewer looking for a free airdrop or passive income because the reward requires real trading volume.

Access, eligibility, and risk. The rules list the EEA, the United States excluding Maine and New York, and other jurisdictions where Kraken services are available. United Kingdom residents are excluded. Reward recipients need a Kraken Pro account in good standing and at least Intermediate KYC verification before distribution; rewards are credited within 30 days after the promotion ends and are not redeemable for cash. 4

Qualifying volume requires real spot trades, so fees and price movement can create losses before any reward is received. The rules prohibit wash trading, self-dealing, coordinated trading, multiple accounts, and VPN circumvention. Kraken can modify, suspend, or terminate the promotion without prior notice. 4

Safe video test. Start with a screen recording of the official rules and the Promotions enrollment flow. Build a spreadsheet that separates the 1,500 QUID threshold, trading fees, price movement, KYC requirement, and possible reward tier. Only include a live trade if the creator already accepts the full market risk and can show the exact order and fee records. Never frame the event as guaranteed profit.

Where to verify. Use the official QUID Trading Challenge rules and the jurisdiction-specific Kraken Terms of Service linked from that page. The rules control if a promotional banner says something different.

Filming priority

Start with ChatGPT's quiz and memory update for the widest audience and the simplest repeatable test.
Choose Gemini 3.7 Flash when the channel can show a real multimodal coding task instead of a model announcement.
Choose Copilot CLI for a terminal workflow where queueing, approvals, and rollback are visible on screen.
Use JetBrains memory and Ollama when the audience can judge hosted versus local model trade-offs.
Cover the QUID challenge only as a rules-and-risk investigation. The public terms make it verifiable, but the trading threshold keeps it separate from a free airdrop story.

Telegram has no current August item in this pass: Telegram's official blog currently lists its July 14, 2026 Communities and Rich Text Editor update as the latest 2026 post. The fresh airdrop leads also lacked a readable primary-source eligibility path and reward schedule, so they do not clear the same verification bar. 8

Before recording, open the linked official page and the live product surface again. Plan, account, region, rollout, credits, KYC, and promotion terms can change independently of the announcement date.

참고 출처

1
ChatGPT release noteshelp.openai.com
2
Gemini API release notesai.google.dev
3
GitHub Copilot weekly releases - August 10github.blog
4
QUID Trading Challenge 2026 official ruleskraken.com
5
Copilot memory and Ollama in GitHub Copilot for JetBrainsgithub.blog
6
Gemini 3.7 Flash model pageai.google.dev
7
GitHub Copilot CLIgithub.com
8
Telegram Blogtelegram.org