State of AI Tools: April 2026

April 2026 was a signal month. Anthropic shipped Claude Opus 4.7, gated Claude Mythos Preview behind Project Glasswing, pushed Claude Cowork to general availability, and the agentic-coding race consolidated around serious IDE and CLI workflows instead of shallow wrappers.

This report preserves the April 2026 snapshot, originally produced when the AiPedia catalog covered 132 tools. The live catalog has since expanded, so the May 13 refresh focused on verifying the volatile April claims and updating the May watchlist with confirmed launches, rather than rewriting the historical issue around today’s count. High-impact claims below are tied to primary or official sources where possible.

June 7 refresh note: This page is now treated as a historical April archive, not a live buyer guide. Current official checks kept the Anthropic Opus 4.7 / Project Glasswing context, confirmed GPT-5.5 as an April 23 OpenAI launch, and rechecked Claude, ChatGPT, Cursor, Midjourney, and ElevenLabs pricing anchors. The May-watchlist section was corrected where current official sources did not support the earlier date claims: Google dates Gemini 3.1 Pro to February 19, 2026, and Mistral dates Mistral 3 to December 2, 2025. Veo 3.1 and Runway Gen-5 stay as watch-list items because the June 7 pass did not find current official sources confirming the claimed April/May launch dates.

May 13 refresh note, superseded by June 7 checks: Opus 4.7, Project Glasswing / Mythos Preview, Claude Cowork GA, Claude Pro, ChatGPT Plus, Cursor Pro, Midjourney Standard, and ElevenLabs Creator pricing were rechecked against May 2026 sources. The May prediction section was updated at the time with a mix of confirmed launches and watch-list items; the June 7 correction above is the current editorial status.

The headline story

On April 16, 2026, Anthropic released Claude Opus 4.7. The model narrowly retook the lead for most powerful generally-available LLM on agentic coding, scaled tool use, computer use, and financial-analysis benchmarks. API sticker price is unchanged from 4.6 at $5 in / $25 out per MTok, but a new tokenizer produces 1.0-1.35× more tokens workloads need to benchmark before migrating.

Nine days earlier, on April 7, Anthropic announced Claude Mythos Preview, a specialised frontier model for defensive cybersecurity, developed under Project Glasswing. The launch partners include AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Anthropic says Mythos Preview has already identified thousands of zero-day vulnerabilities across critical infrastructure and is available only as a gated research preview.

Mythos is invitation-only. No self-serve API, no Max-tier access, no public waitlist. A major US lab has pulled a second frontier model out of its research pipeline and gated it behind a defence consortium, which is a first for this generation of commercial frontier AI deployment.

Practical effect for anyone building on Claude: the main lineup is now Opus 4.7, Sonnet 4.6, and Haiku 4.5, all accessible through claude.ai, the API, and Claude Code. Mythos is not replacing Opus and is not coming to Pro.

The second-order question is whether this becomes a pattern. If OpenAI or Google respond with their own gated frontier models for defence or biosecurity, the top of the capability curve splits into public and private tracks. That would change how the aipedia.wiki moat score works across the lab tier. For now, one data point.

New and noteworthy

Claude Desktop and Claude Cowork reached general availability on macOS and Windows on April 9, 2026. The desktop app is serviceable. The Cowork feature (shared Claude sessions across multiple humans) is the more interesting of the two, because it solves the “everyone has their own ChatGPT history” problem that every team using AI assistants has hit. Small teams at small companies are the audience; large orgs already run Enterprise.

Computer Use inside Claude Code is in research preview as of the week of March 30 to April 3, 2026. It is exactly as rough as a research preview suggests. On a simple “open Chrome, find this page, scroll to pricing, copy the numbers” task, it works. Latency is 2 to 4 seconds per action and error recovery is poor. Not production-ready. Teams building agentic workflows should start experimenting now.

Claude Code’s April agent features became clearer after launch. Anthropic’s Week 15 Claude Code notes covered the Monitor tool and Ultraplan cloud planning, and the current commands documentation still lists /ultraplan and /ultrareview. The April 14 v2.1.108 package release added recap and cache-control improvements rather than being the original source of Monitor and Ultraplan.

Cursor 2.0 is the current stable release. The delta versus Cursor 1.x is noticeable for users coming off older versions, unremarkable for active daily users. It remains the leading GUI agentic IDE. Claude Code is eating its coding-hour share for multi-file sessions.

Qwen3.6 Plus with 1M context is the open-weight story still worth watching. Released February 2026, taking longer than usual to shake out in the community because the inference costs on the MoE architecture surprised users in both directions. Cheaper than expected on small tasks, slightly more expensive than expected on very long contexts.

Score changes

Three score movements this month. Each is editorial, not data-driven.

• Claude moved from 9.5 to 9.6. Opus 4.7 takes measurable leads on agentic coding, tool use, computer use, and financial analysis. Tokenizer-driven cost bump holds the move to 0.1 rather than 0.2.
• Claude Code held at 9.5. Ultraplan and Monitor are useful; Computer Use preview is ahead of schedule. Sharp edges for beginners remain, and research previews do not count toward scoring. Revisit in May.
• Sora (dead) moved to the archive with a final score of 7.2. Cleanup from the March discontinuation. The final score reflects the product as it shipped, not a post-mortem.

No other tool in the catalog had a material change this month that would justify a score move. Scores hold until something real happens.

Pricing shifts

No major pricing shifts confirmed across the catalog in April 2026. Claude Pro held at $20/mo, ChatGPT Plus held at $20, Cursor Pro held at $20, Midjourney Standard held at $30, ElevenLabs Creator held at $22.

The Claude API sticker is unchanged on Opus 4.7 ($5 in / $25 out per MTok), Sonnet 4.6 ($3 / $15), and Haiku 4.5 ($1 / $5). The Opus 4.7 tokenizer bump is the only effective-cost shift; see Claude tool page for the detail.

The graveyard

No new confirmed deaths in April 2026. Most recent confirmed discontinuations: Sora (March 2026), GPT-4o retired by OpenAI on February 13, 2026, DALL-E retired March 2025. Inflection’s Pi was already in maintenance mode before April and does not count as an April death.

Watching for late-month or early-May confirmation: whether any of the smaller Chinese open-weight projects stop pre-training (following the pattern 01.AI set in March 2025) and whether any of the “AI browser” wrapper startups from the 2024 cohort quietly shut down. No public obituaries yet.

What the May watchlist looked like (June 7 status)

Claude Sonnet 4.7. Still unconfirmed by the June 7 refresh. Anthropic’s public release notes list Opus 4.7 and Claude Design in April, and the May 9 to 12 burst was enterprise (FIS, Akamai, Claude for Legal) rather than a Sonnet point release. Keep on the watch list.

Runway Gen-5. Still unconfirmed by the June 7 refresh. The video market remains led by Seedance, Kling, Veo, Runway, Pika, Luma, and Hailuo lanes on current buyer pages, but this archived report should not be read as confirming a Gen-5 launch.

Gemini 3.2. Not released as of the June 7 refresh. Google instead kept the Gemini 3.1 line current, but the official Gemini 3.1 Pro announcement is dated February 19, 2026, not April 16. Treat Gemini 3.2 as speculation; do not use this archived report as a launch-date source for Gemini.

Confirmed and corrected follow-ups. GPT-5.5 remains confirmed from OpenAI’s April 23 announcement, with API availability noted on April 24. Mistral 3 was removed from the April 28 follow-up list because Mistral’s official page dates the family to December 2, 2025. Daybreak, Deployment Company, Codex Chrome, Claude for Legal, Gemini Intelligence / Googlebook, Isomorphic Labs funding, and the reported Anthropic financing remain archived watch-list items unless a linked current source confirms them elsewhere on AiPedia.

Agentic coding consolidation. Claude Code, Codex (now with Chrome), and Aider anchor the CLI agentic space. Cursor still owns the GUI. The mid-market (“agentic but in VS Code as an extension”) is a graveyard in slow motion. Fewer than two extension-based tools are likely to survive into 2027.