AI This Week: Sonnet 4.6, Gemini 3.1 Pro, and seven models in one month

Seven frontier models in one month. That's where February 2026 stands as I write this. The releases keep coming, and the gap between open-source and proprietary performance is now close to zero on most benchmarks. Here's what happened this week.

Top stories this week

Sonnet 4.6 delivers Opus-class performance at Sonnet prices

Anthropic released Claude Sonnet 4.6 on February 17th. It scores 79.6% on SWE-bench Verified and 72.5% on OSWorld. For comparison, Opus 4.6 sits at 80.8% on SWE-bench and 72.7% on OSWorld. That's a 1.2 percentage point gap on coding and essentially a tie on computer use.

The pricing makes this interesting. Sonnet 4.6 costs $3 per million input tokens and $15 per million output tokens - five times cheaper than Opus. It also gets a 1 million token context window in beta, a first for any Sonnet-class model.

Anthropic has made it the default model in claude.ai and Claude Cowork. In Claude Code testing, developers preferred Sonnet 4.6 over Sonnet 4.5 70% of the time and over the previous flagship Opus 4.5 59% of the time. That last number is the one that should worry anyone selling premium-tier models: when the cheaper model wins most head-to-head comparisons with the more expensive one, the pricing structure becomes hard to defend.

I've been using it since launch. For most of my coding tasks, the quality difference with Opus 4.6 is imperceptible. The cost difference is not. This is the kind of release that changes what model you reach for by default.

Gemini 3.1 Pro doubles reasoning performance

Google released Gemini 3.1 Pro on February 19th. The headline: 77.1% on ARC-AGI-2, which is designed to test whether a model can solve entirely new logic patterns it hasn't seen during training. That's roughly double the previous Gemini 3 Pro's score, about 24% ahead of GPT-5.2, and nearly 9% ahead of Opus 4.6.

On coding, it holds its own too: 80.6% on SWE-bench Verified, which puts it at the top of the leaderboard alongside Opus 4.6. Pricing sits at $2 per million input tokens and $12 per million output tokens for prompts under 200K tokens, essentially matching the previous Gemini 3 Pro rates. A free upgrade for existing users.

ARC-AGI-2 is the number to pay attention to here. Reasoning benchmarks have been a persistent weakness for most models. Doubling performance on a benchmark specifically designed to resist memorization-based shortcuts suggests Google has made a genuine architectural or training improvement, not just scaled up.

Windsurf Wave 13 ships parallel multi-agent coding

Windsurf released Wave 13 with three notable additions: their SWE-1.5 model is free for all users through March, parallel multi-agent sessions via Git worktrees, and a multi-pane interface for monitoring multiple agents side by side.

The Git worktrees integration is the real feature here. You can run five Cascade agents on five different bugs simultaneously, each on a different branch sharing Git history but without file conflicts. Each agent gets its own dedicated zsh terminal with environment variable inheritance.

This is Windsurf's answer to Claude Code's Agent Teams, which shipped with Opus 4.6 earlier this month. Different implementation, similar idea: coding work increasingly looks like managing a team of agents rather than writing code yourself. The tools are converging on multi-agent orchestration as the default paradigm.

Free SWE-1.5 for three months is a smart acquisition play. Windsurf is betting that once developers experience their coding model, the switching cost keeps them around when the free period ends.

Claude Code Security crashes cybersecurity stocks

On February 20th, Anthropic announced Claude Code Security, a research preview that scans codebases for vulnerabilities and suggests patches for human review. Using Opus 4.6, Anthropic's team found over 500 vulnerabilities in production open-source codebases, bugs that had gone undetected despite years of expert review.

The market panicked. CrowdStrike dropped 7.8%, Cloudflare slid 5.9%, Palo Alto Networks declined 6.4%. Smaller static-analysis vendors like JFrog fell nearly 25%. Billions in market cap evaporated in hours.

The reaction was disproportionate. Claude Code Security is static analysis. It reads code and reasons about vulnerabilities. It does not test runtime behavior, cannot send requests through an API stack, and can't confirm whether a finding is exploitable in a live environment. It's a code reviewer, not a penetration tester.

That matters because most of the companies that got hit sell runtime security, endpoint protection, and network monitoring. Claude Code Security doesn't compete with CrowdStrike's endpoint detection or Cloudflare's DDoS mitigation. It competes with Snyk, SonarQube, and manual code audit. The market sold the wrong stocks.

Still, the signal is real. AI-driven code analysis that reasons about vulnerability patterns rather than matching known signatures is a meaningful advance. Traditional SAST tools will need to evolve. But the cybersecurity industry as a whole isn't under threat from a code scanner.

The February model rush: seven frontier models in one month

Step back and count: Opus 4.6 (February 5), GLM-5 (February 11), Qwen3.5 (February 16), Sonnet 4.6 (February 17), Gemini 3.1 Pro (February 19), with DeepSeek V4 and others still expected before month's end. Seven frontier-class models in a single month.

Three patterns stand out.

The open-closed gap is near zero. GLM-5 (MIT) scores 77.8% on SWE-bench. Qwen3.5 (Apache 2.0) runs at 60% less than its predecessor. Sonnet 4.6 (proprietary but cheap) delivers Opus-class performance at Sonnet pricing. The premium for frontier proprietary models is shrinking fast.

Pricing is in free fall. When multiple frontier models compete for the same workloads, the price floor drops. Qwen3.5 at $0.18 per million input tokens, Sonnet 4.6 at $3, Gemini 3.1 Pro at $2 - these numbers would have been unthinkable six months ago.

Model routing becomes essential. No single model wins everything. Gemini 3.1 Pro leads on reasoning (ARC-AGI-2). Opus 4.6 leads on agentic coding (SWE-bench). GLM-5 offers the best cost-performance on Huawei hardware. Qwen3.5 dominates multilingual coverage. The teams that build routing logic to select the right model per task will have a structural cost advantage.

What does this actually mean?

February 2026 will be remembered as the month the frontier stopped being exclusive. A year ago, frontier performance meant paying premium pricing to one of three US companies. Today, you can get there with an MIT-licensed Chinese model, an Apache 2.0 Chinese model, a European model in a Swedish data center, or a cheaper Anthropic model. The options exist.

The practical shift for builders: stop thinking about "the best model" and start thinking about model portfolios. Route reasoning tasks to Gemini 3.1 Pro, coding to Opus 4.6, multilingual to Qwen3.5, cost-sensitive to Sonnet 4.6 or GLM-5. Treat model selection like infrastructure selection: use the right tool for each job.

The cybersecurity market reaction to Claude Code Security is a preview of what happens next. Every time AI gets credibly good at a professional task, the market for legacy tooling in that space gets repriced. Static analysis tools are first. Runtime security, testing, monitoring will follow. None of it means replacement, but the tools that don't evolve will lose market share to the ones that do.

What to watch next week

• DeepSeek V4 is expected any day: Apache 2.0, trillion-parameter MoE, 1M context. If it delivers, it adds another viable open-source frontier option.
• Claude Code Security in practice: I'm running it on some of my projects this week. Interested to see how it handles real-world codebases versus the cherry-picked examples in the announcement.
• February's final tally: we should have a clear count of frontier releases by month's end. Whether it's seven or eight, it's already unprecedented. March is going to look quiet by comparison.