DeepSeek's Flagship Chatbot Gains Vision for the First Time
If you have been relying on DeepSeek for text-based research or coding tasks, the platform changed shape significantly on 29 April. DeepSeek quietly rolled out an image recognition mode to its flagship chatbot in a grey-scale beta test, making it the first time the company has opened multimodal capability directly to consumer users. The new mode sits alongside the existing Expert and Flash chat options, allowing users to submit images for analysis and reasoning.
What makes this technically notable is that the V4 model underneath was built with a native multimodal architecture integrated during pre-training, rather than a vision module bolted on after the fact. A model trained to understand images from the ground up handles cross-modal reasoning more naturally than one that treats vision as a plugin. The rollout was first signalled via a post on X by Chen Xiaokang, head of DeepSeek's multimodal team, who shared an image of the company's signature blue whale with one eye now open, ahead of mainstream press coverage.
1M Token Context Window Beta Retires Today for Claude Sonnet 4 and Sonnet 4.5
If your team is running API calls to Claude Sonnet 4 or Sonnet 4.5 with the extended context beta header, today is the hard cutoff. From 30 April 2026, the beta header anthropic-beta: context-1m-2025-08-07 is no longer honoured on those two models, and any prompt that exceeds 200,000 tokens will return a 400 error. This is not a model deprecation: Sonnet 4 and Sonnet 4.5 continue to run with a standard 200k context window until June 2026.
Anthropic's recommended migration path is straightforward: switch to Claude Sonnet 4.6 or Opus 4.6, both of which now ship the 1M token context window as a generally available feature with no additional beta header and no surcharge. Any pipeline currently processing large codebases, lengthy documents, or extended conversation histories in a single prompt that has not been updated before today will begin returning errors immediately.
Bloomberg reported on 29 April that Anthropic is weighing a fresh funding round that would value the company at more than $900 billion, potentially making it the world's highest-valued AI startup. No offers have been accepted; the consideration is described as at an early stage. Bloomberg, 29 Apr
Comet AI Browser Receives a Significant iPad Update
If you use Perplexity's Comet browser on iPad, a notable update published on 29 April brings a refined interface optimised for larger screens, making it considerably easier to multitask between AI-powered research and standard web browsing. Comet embeds an AI assistant directly into the browser layer, enabling in-page summarisation, autonomous multi-step tasks, and research without leaving the browser. The iPad-specific refinements suggest Perplexity is treating tablet users as a meaningful segment of its power user base rather than a secondary port.
This update follows Comet's recent enterprise tier launch, which added MDM deployment, granular agent action controls, and full audit logging for IT administrators, positioning the browser as a managed enterprise tool as well as a consumer productivity product. The direction of travel is clear: Comet is being built as a full AI-native browser for both individual and organisational use, not a novelty experiment.
Azure Cloud Revenue Up 40% in Q3 FY2026, Driven by AI Infrastructure Demand
Microsoft's third-quarter FY2026 results, published on 29 April, show Azure and other cloud services revenue growing 40% year-on-year, a figure the company attributed directly to surging demand for AI infrastructure and model hosting. Overall revenue came in at $82.9 billion, up 18% year-on-year.
For teams evaluating AI platform decisions, sustained Azure growth at this scale signals that Microsoft is continuing to invest heavily in AI compute capacity rather than pulling back after a headline-driven spike. Notably, the quarter covers the period immediately following Microsoft's partnership restructure with OpenAI, which saw OpenAI's models made available on competing cloud platforms for the first time. Azure's numbers show no visible deterioration from that change, suggesting the structural demand for AI infrastructure on Azure is independent of OpenAI exclusivity.
Industry Themes
Three patterns are visible across today's news. The first is that multimodal capability is becoming a competitive baseline rather than a differentiator. DeepSeek's rollout of image understanding to consumer users, surfaced first via the company's own X account before mainstream coverage, signals that even cost-focused, open-weight model providers now feel pressure to match the multimodal standards set by GPT-4o and Gemini. When a company positioned primarily on price and openness adds vision, it is no longer a premium feature: it is table stakes.
The second is that API deprecation cycles are tightening. Anthropic's retirement of the extended context beta today, just months after the feature was introduced, is a reminder that building production pipelines on top of beta features carries real operational risk. Developers should factor shorter shelf lives into any pipeline that depends on experimental model capabilities, and treat migration planning as a standing maintenance task rather than a one-off project.
The third is that enterprise cloud spend on AI infrastructure continues to accelerate. Microsoft's 40% Azure growth in the quarter ending March 2026 suggests AI investment is translating into durable infrastructure spend rather than remaining a pilot-stage phenomenon, and that the structural demand is not dependent on any single vendor's model exclusivity. For anyone modelling AI infrastructure costs for the year ahead, the directional signal from these results is that cloud AI capacity pricing is unlikely to soften in the near term.