Wiredframe Radar logo

Category intelligence

Social Media Briefing — April 25, 2026

527 current items analyzed and ranked.

Executive synthesis

Social Media Summary

Two massive launches dominated AI discourse: OpenAI's GPT-5.5 API release and DeepSeek V4's open-source drop. Sam Altman announced GPT-5.5 and GPT-5.5 Pro API availability, while Perplexity immediately swapped Claude Opus 4.7 for GPT-5.5 as its default orchestrator — a notable market signal.

  • DeepSeek V4 drew the most technical analysis: vLLM shipped day-0 support with a deep dive into the novel attention mechanism, while Scobleizer highlighted V4-Pro at $3.48/M output tokens — roughly 1/5th the cost of Claude Opus 4.6 or GPT-5.4
  • Ethan Mollick and Simon Willison provided hands-on evaluations of DeepSeek V4; Mollick called it 'surprisingly only okay' and ranked it behind Kimi K2.6, tempering benchmark hype
  • Yann LeCun sparked major debate (233K views) arguing LLMs cannot deliver a robot-rich future and that Joint Embedding Architectures will surpass generative approaches
  • Anthropic unveiled Project Deal, a novel internal marketplace where Claude negotiated on behalf of employees — finding that more capable models secured better deals invisibly
  • Cohere and Aleph Alpha announced a transatlantic merger focused on sovereign enterprise AI, signaling growing geopolitical fragmentation in AI infrastructure

Key Themes

GPT-5.5 Launch & API Availability · 14DeepSeek V4 Launch & Infrastructure · 4DeepSeek V4 Launch · 10GPT-5.5 API Launch · 6AI Pricing Competition · 4DeepSeek V4 Release and Reception · 16LLM Limitations & JEPA/World Models (LeCun) · 14Anthropic Project Deal: AI Agent Marketplace Experiment · 11OpenAI Codex & Agent Platform Evolution · 10DeepSeek-V4 Release · 3

Primary evidence

Top Ranked Signals

Social Twitter Apr 24

🎉 Day-0 support for @deepseek_ai V4 Pro and Flash on vLLM — a new generation of DeepSeek model, purp...

By @vllm_project

92 score
AI Analysis

Following yesterday's Reddit release announcement, vLLM project announces day-0 support for DeepSeek V4 Pro and Flash with detailed technical breakdown of the new attention mechanism: shared K/V with inverse RoPE (2x memory savings), multi-level KV compression (4x-128x savings), sparse attention, optimized kernels with 1.4-20x speedups.

🎉 Day-0 support for @deepseek_ai V4 Pro and Flash on vLLM — a new generation of DeepSeek model, purpose-built for tasks up to 1M tokens. Alongside the release, we're publishing a first-principles walkthrough of the new long-context attention and how we implemented it in vLLM. The new attention mechanism, in four moves: • Shared K/V + inverse RoPE → 2× memory savings • c4a / c128a KV compression → 4×–128× savings • DeepSeek Sparse Attention over compressed tokens • Short sliding window for loca
DeepSeek V4 launchvLLMInference optimizationAttention mechanismsKV cache optimizationOpen source infrastructure
Social Twitter Apr 24

DeepSeek 4 is out. My AI says: +++++ The Numbers That Matter V4 Pro costs $3.48 per million outp...

By @Scobleizer

92 score
AI Analysis

Following yesterday's Reddit release announcement, Scobleizer provides a comprehensive summary of the DeepSeek V4 launch: two open-source models (V4-Pro 1.6T params, V4-Flash 284B params), both with 1M context. V4-Pro costs $3.48/M output tokens vs Claude Opus 4.6 at $25 and GPT-5.4 at $15. ValsAI ranks V4 #1 on Vibe Code Benchmark. Includes reactions from industry figures, Sam Altman's cryptic response about GPT-5.5, and skeptical takes about benchmaxxing.

DeepSeek 4 is out. My AI says: +++++ The Numbers That Matter V4 Pro costs $3.48 per million output tokens. Claude Opus 4.6 costs $25. GPT-5.4 costs $15. Same benchmark tier. One fifth the price. ValsAI ran independent tests. V4 is now number 1 on their Vibe Code Benchmark. Not just among open models. Among all models. It beats Gemini 3.1 Pro. t.co/W69DdrX22K is updated with 26 new stories. Essay: "China Just Fired Back. And This Time the Benchmarks Are Real." Here's the summary
DeepSeek V4 LaunchAI Pricing CompetitionOpen Source AIBenchmarksUS-China AI Competition
Social Twitter Apr 24

@ben_j_todd 1. I never said LLMs were not useful. They are, particularly with all the bells and whis...

By @ylecun

88 score
AI Analysis

Following yesterday's News coverage of LeCun's AMI Labs, Yann LeCun responds to a cartoon about AI's future: (1) LLMs are useful but won't enable a robot-rich future since they don't understand the physical world, (2) world models and zero-shot planning are needed, plus humorous personal notes about not wearing ties.

@ben_j_todd 1. I never said LLMs were not useful. They are, particularly with all the bells and whistles that are being added to them. I use them. 2. A robot-rich future can't be built with AIs that don't understand the physical world and don't anticipate the consequences of their actions. And LLMs really don't. 3. The future in the cartoon looks pretty dystopian TBH, but even a non-dystopian version will require world models and zero-shot planning abilities. 4. I rarely wear a suit and absolut
LLM limitationsroboticsworld modelsphysical understandingAI futureJEPA
Social Twitter Apr 24

While everyone watched GPT-5.5 launch, DeepSeek quietly shipped V4 the next morning. V4-Pro: 1.6T t...

By @AlphaSignalAI

88 score
AI Analysis

Following yesterday's Reddit release announcement, AlphaSignalAI provides a detailed breakdown of DeepSeek V4 launch: V4-Pro (1.6T/49B active) and V4-Flash (284B/13B active), MIT licensed, native 1M-token context, 8.6x cheaper than GPT-5.5, with top coding benchmarks including LiveCodeBench 93.5 and Codeforces 3206.

While everyone watched GPT-5.5 launch, DeepSeek quietly shipped V4 the next morning. V4-Pro: 1.6T total / 49B active, MIT license. V4-Flash: 284B total / 13B active. Both with native 1M-token context. At 1M tokens, V4-Pro runs at 27% of V3.2's FLOPs and 10% of the KV cache. Pricing: $3.48 per 1M output tokens. 8.6× cheaper than GPT-5.5. 7.2× cheaper than Opus 4.7. Coding benchmarks: • LiveCodeBench: 93.5 (beats Gemini 3.1 Pro, Opus 4.6) • Codeforces: 3206 (highest in the frontier compari
DeepSeek V4 launchModel benchmarksAI pricing competitionOpen-weight modelsLong context
Social Bluesky Apr 24

DeepSeek V4 just dropped - two models, Flash and Pro, both benchmarking well, decent pelicans and pr...

By @simonwillison.net

88 score
AI Analysis

Following yesterday's Reddit release announcement, Simon Willison covers the DeepSeek V4 launch, noting both Flash and Pro models benchmark well, have decent pelicans (his standard test), and are the cheapest in their respective categories by a solid margin.

DeepSeek V4 just dropped - two models, Flash and Pro, both benchmarking well, decent pelicans and prices that put them both as the cheapest in their respective categories by a solid margin simonwillison.net/2026/Apr/24/...
DeepSeek V4 LaunchAI Model EvaluationAI Pricing Competition
Social Bluesky Apr 24

My first two TiKZ Sparks unicorns from DeepSeek v4. Um. (Expert mode, from the DeepSeek site, which...

By @emollick.bsky.social

85 score
AI Analysis

Following yesterday's Reddit release announcement, Ethan Mollick tests DeepSeek V4 Pro using his TiKZ Sparks unicorn benchmark and finds the results underwhelming ('Um.'). His follow-up post says V4 is 'surprisingly only okay' and behind Kimi K2.6 in his opinion.

My first two TiKZ Sparks unicorns from DeepSeek v4. Um. (Expert mode, from the DeepSeek site, which is supposed to be v4 Pro according to the release)
DeepSeek V4 LaunchAI Model EvaluationBenchmarks vs Reality
Social Twitter Apr 24

this has one of the most exciting launch weeks in OpenAI's history, with a goal of making agents mor...

By @gdb

82 score
AI Analysis

Greg Brockman calls this one of the most exciting launch weeks in OpenAI history, summarizing Codex improvements: smarter computer use, better context retention, more independent agent operation.

this has one of the most exciting launch weeks in OpenAI's history, with a goal of making agents more real, useful, and accessible for all our users. codex can now smartly do much more on your computer, remember more of your context, and run more ongoing work independently.
OpenAI launch weekAI agentsOpenAI Codexproduct strategy
Social Twitter Apr 24

auto-review now live in codex — using a guardian agent to evaluate the safety of proposed actions, r...

By @gdb

82 score
AI Analysis

Greg Brockman announces 'auto-review' feature in Codex — a guardian agent that evaluates proposed action safety, reducing unnecessary human approval steps.

auto-review now live in codex — using a guardian agent to evaluate the safety of proposed actions, reducing human approvals to only when they're really needed.
OpenAI CodexAgentic AIAI SafetyProduct Launch
Social Twitter Apr 24

New Anthropic research: Project Deal. We created a marketplace for employees in our San Francisco o...

By @AnthropicAI

82 score
AI Analysis

Anthropic announces Project Deal: an internal marketplace experiment where Claude bought, sold, and negotiated on behalf of employees in their SF office.

New Anthropic research: Project Deal. We created a marketplace for employees in our San Francisco office, with one big twist. We tasked Claude with buying, selling and negotiating on our colleagues’ behalf. t.co/H2f6cLDlAW
AI Agent MarketsAnthropic ResearchProject DealAI Commerce
Social Twitter Apr 24

i'm a few days late to realizing this but: wow, opus 4.7 is god awful like so, so bad it's making...

By @mattshumer_

82 score
AI Analysis

Matt Shumer strongly criticizes Claude Opus 4.7 as 'god awful', saying it makes mistakes he'd expect GPT-4o to handle, and asks if there's an explanation. Massive engagement (1,031 likes, 134K views).

i'm a few days late to realizing this but: wow, opus 4.7 is god awful like so, so bad it's making mistakes on things i'd expect gpt-4o to handle cleanly there's got to be some explanation, right?
Claude Opus 4.7Model quality regressionAnthropic criticismModel evaluation