The week of April 16 to 26, 2026 marks a notable evolution for AI-assisted development tools. OpenAI turns Codex into a versatile workspace with an integrated browser, macOS computer use, and the arrival of GPT-5.5 as the recommended model. GitHub strengthens Copilot’s integration with Jira through four new enterprise capabilities. DeepSeek announces a permanent and immediate reduction in input cache pricing — cut to 1/10th across the entire API series.
Codex CLI: integrated browser, macOS computer use, and GPT-5.5
April 16 and 23, 2026 — Over two weeks, OpenAI releases four versions of Codex CLI and repositions the tool. Version v0.122.0, released on April 16, lays the groundwork for a “broader workspace” (broader workspace); version v0.125.0, released on April 23, adds GPT-5.5 as the recommended model.
“Codex is becoming a broader workspace for getting work done with AI. This update makes it easier to start work with less setup, verify what Codex is building, create richer outputs, and keep momentum across longer-running tasks.” — Codex Changelog v0.122.0
Integrated browser and computer use
Version v0.122.0 introduces two new interface perception capabilities. The integrated browser makes it possible to open and annotate local or public web pages directly from the Codex app, without authentication. Users can visually mark elements of the rendered page and ask Codex to process this feedback.
The computer use feature extends this logic to native macOS applications: Codex can see the screen, click, and type text. This covers application testing, mobile simulator flows, and graphical interface debugging. This feature is subject to a geographic restriction at launch: it is not available in the European Economic Area (EEA), the United Kingdom, or Switzerland.
In addition, projectless chats make it possible to start an AI conversation without first selecting a working directory — useful for research, writing, or exploratory tasks. Enhanced outputs now make it possible to generate diagrams, tables, and exportable files in responses.
GPT-5.5 as the recommended model in Codex
Version v0.125.0 integrates GPT-5.5 as the main model in the Codex ecosystem. It is presented as the recommended choice for implementation, refactoring, debugging, testing, and validation. Selection is done via codex --model gpt-5.5 from the command line, or from the model picker in the IDE or app. If GPT-5.5 does not yet appear in the interface, an update to the latest version is required.
Summary of the 4 versions (April 16-23)
| Version | Date | Content |
|---|---|---|
| v0.122.0 | April 16 | Integrated browser, macOS computer use (outside EEA/UK/CH), projectless chats, enhanced outputs |
| v0.123.0 | April 20 | Technical release — no new user-facing feature |
| v0.124.0 | April 23 | Technical release — no new user-facing feature |
| v0.125.0 | April 23 | GPT-5.5 recommended model in CLI, IDE, and Codex app |
🔗 Codex CLI v0.122.0 — GitHub Release · Codex CLI v0.125.0 — GitHub Release
GitHub Copilot for Jira: 4 new enterprise capabilities
April 22, 2026 — GitHub enriches the Copilot cloud agent integration with Jira. Four new capabilities are available, all focused on finer-grained control and better alignment with existing team conventions.
| Capability | Description |
|---|---|
| Custom agents | Specify in the Jira ticket which custom agent (from the GitHub repository) should handle the task |
| Atlassian custom fields | Copilot reads acceptance criteria and other custom fields to enrich its working context |
| Branch rules | The agent follows the branch naming conventions defined in the ticket, ensuring consistency when creating pull requests |
| Global instructions | Define at the Atlassian space level a target repository, branch rules, specific agents — automatically applied at each trigger |
Together, these capabilities give teams granular control over Copilot’s behavior in Jira workflows, without having to manually reconfigure each interaction.
In addition, an automatic notification is now posted in the corresponding Jira ticket when Copilot opens a draft pull request and requests a review. This keeps the team informed directly in Jira, without having to monitor GitHub in parallel.
🔗 GitHub Changelog — Copilot for Jira enhancements
DeepSeek: input cache price cut by 10
April 26, 2026 — DeepSeek announces a permanent and immediate reduction in the price of input cache hits across the entire DeepSeek API series. The cost is reduced to 1/10th of the original price.
DeepSeek Input Cache Price Drop! Effective immediately, the price for input cache hits across the ENTIRE DeepSeek API series is reduced to just 1/10th of the original price! Build more efficiently for less. Reminder: The DeepSeek-V4-Pro 75% OFF promotion is still active until May 5th, 2026, 15:59 (UTC Time).
The reduction covers all models in the lineup: V4-Pro, V4-Flash, and earlier versions. It is permanent and separate from the temporary -75% promotion on V4-Pro (still active until May 5, 2026 at 15:59 UTC).
| Parameter | Detail |
|---|---|
| Reduction | ×10 on input cache hit pricing |
| Scope | Entire DeepSeek API series |
| Effect | Immediate |
| V4-Pro -75% promo | Still active, expires May 5, 2026 at 15:59 UTC |
The announcement was viewed more than 105,000 times on X.
Briefs
-
Codex CLI v0.123.0 + v0.124.0 — Technical releases published on April 20 and April 23 respectively, with no new user-facing features. 🔗 v0.123.0 · 🔗 v0.124.0
-
NotebookLM — Automatic source categorization (April 24) — As soon as there are 5 sources in a notebook, NotebookLM can label and group them automatically. Labels are customizable (rename, emojis). Gradual rollout. 🔗 @NotebookLM on X
-
NotebookLM — Simultaneous multi-email sharing (April 24) — No more adding email addresses one by one to share a notebook: you can now paste a list of addresses and send in one step. 🔗 @NotebookLM on X
-
GitHub Copilot SDK — React Native tutorial (April 25) — Article by @acolombiadev (GitHub Developer Advocate) on integrating the Copilot SDK into a React Native app: AI-generated GitHub issue summaries, with production patterns (graceful degradation, cache). 🔗 GitHub Blog
What this means
These announcements point to a coherent shift: AI development tools are becoming more deeply integrated into existing workspaces, rather than remaining parallel tools. Codex no longer just writes code — it can now browse the web, interact with macOS applications, and produce enhanced outputs, while receiving the best model available from OpenAI. GitHub Copilot for Jira follows the same logic: instead of imposing a workflow change, it adapts to the conventions and tools already in place within teams.
DeepSeek’s price cut on cache hits fits into a broader pricing pressure dynamic that directly benefits teams building large-scale applications. Caching is a central lever for controlling costs in agent pipelines — reducing this item to 1/10th changes the economic equation for many projects.
Sources
- Codex Changelog — OpenAI
- Codex CLI v0.122.0 — GitHub Release
- Codex CLI v0.125.0 — GitHub Release
- GitHub Copilot for Jira enhancements — GitHub Changelog
- Building AI-powered GitHub issue triage with the Copilot SDK
- Tweet @deepseek_ai — Input Cache Price Drop
- Tweet @NotebookLM — Auto-labeling sources
- Tweet @NotebookLM — Multi-email sharing
- Tweet @github — Copilot SDK React Native
This document was translated from the fr version into en using the gpt-5.4-mini model. For more information about the translation process, see https://gitlab.com/jls42/ai-powered-markdown-translator