The AI Wire ●

OpenAI released GPT-5.3-Codex, a return to code-specialized fine-tuning after years of generalist flagships. The agentic coding model hit 56.8% on SWE-bench Pro and 77.3% on Terminal-Bench 2.0, running about 25% faster on tool-driven tasks.

The name was a deliberate callback to the original 2021 Codex, but the model was a different beast: agent-optimized rather than a simple code completion engine. It reached GA in Copilot on February 9 across Pro, Business, and Enterprise tiers.

OpenAI later designated GPT-5.3-Codex its first long-term-support model, guaranteeing it for Business and Enterprise Copilot users through February 2027.

Anthropic released Claude Opus 4.6 with a 1-million-token context window and support for agent teams — multiple coordinated sub-agents working under one orchestrator. Claude Sonnet 4.6 followed on February 17 as the everyday workhorse tier.

The 1M context put Claude on par with the longest-context rivals, and agent teams formalized the multi-agent patterns developers had been hand-rolling. Anthropic framed the release around autonomous teams rather than single agents.

Opus 4.6 held the flagship slot through the first half of 2026, until the Fable line and Opus 5 arrived later in the year.

Google released Gemini 3 Flash, the speed-optimized sibling of Gemini 3 Pro, and made it the default model in the Gemini app for most users. It delivered near-Pro quality on everyday tasks at much lower latency.

The Flash tier's promotion to default showed how far Google's efficiency push had come: most users never needed the heavier Pro model. Flash became the volume workhorse of the Gemini lineup.

The release kicked off an accelerating Flash cadence in 2026, with Google shipping new Flash iterations roughly monthly while the next Pro flagship slipped.

OpenAI released GPT-5.2, a three-mode family — Instant, Thinking (with standard and extended effort), and the heavier GPT-5.2 Pro — with gains in instruction following, vision, front-end code generation, and token efficiency. A new xhigh reasoning level and context compaction arrived for long-running work.

The launch came about three weeks after Google's Gemini 3 Pro seized the multimodal lead, and reporting described an internal 'Code Red' memo accelerating GPT-5.2's release. OpenAI executives pushed back on the reactive framing while acknowledging the extra resources helped.

A coding-specialized GPT-5.2-Codex followed on December 18. GPT-5.2 was eventually discontinued in ChatGPT on June 12, 2026.

Anthropic released Claude Opus 4.5, the first model to break 80% on SWE-bench Verified with a score of 80.9%. The launch landed in the middle of a famous two-week stretch in which OpenAI, Google, and Anthropic each shipped flagship coding models days apart.

Opus 4.5 paired the benchmark lead with stronger reasoning and agentic behavior, and Anthropic pitched it as the model for the hardest engineering work. It arrived just as Google's Gemini 3 Pro was resetting expectations elsewhere.

The 80% SWE-bench milestone became a symbolic line in the coding-model race, and Opus 4.5 held it first.

Google launched Gemini 3 Pro, the first model to break 1500 Elo on LMArena, with a 'Deep Think' reasoning mode for the hardest problems. It took the outright frontier lead across benchmarks and public preference rankings.

The release reportedly triggered a 'code red' inside OpenAI, which scrambled its roadmap and rushed GPT-5.2 out within weeks to answer it. The November–December 2025 stretch became one of the most intense head-to-head periods in AI history.

Gemini 3 Pro cemented Google's return to the frontier after years of playing catch-up, and its Deep Think mode became the template for maximum-effort reasoning tiers.

xAI released Grok 4.1, which topped agentic benchmarks and sharply cut hallucinations — down to 4.22% from 12.09% on xAI's evaluations. The update also brought a warmer, more personable default personality.

Grok 4.1's hallucination reduction addressed one of the line's longest-standing weaknesses. xAI pitched it as the model that finally paired Grok's real-time X awareness with trustworthy answers.

The release kept xAI competitive through the intense November 2025 three-lab shootout, holding the line while rivals shipped their own flagships within days.

OpenAI shipped GPT-5.1 as a ChatGPT update with two modes — GPT-5.1 Instant and GPT-5.1 Thinking — and, more unusually, let users select alternative assistant personalities including 'friendly,' 'efficient,' and 'cynical.'

The release also improved coding steerability and reduced the 'overthinking' problem where reasoning models wasted compute on simple questions.

A coding variant, GPT-5.1-Codex-Max, followed on November 18, foreshadowing OpenAI's return to specialized coding models.

Anthropic released Claude Sonnet 4.5, elevating the Sonnet tier to flagship status for agents, coding, and computer use. It was the strongest model in the world for building complex agents at launch, according to Anthropic's evaluations.

Sonnet 4.5 emphasized sustained autonomous work: better planning, more reliable tool use, and stronger performance on long-horizon tasks. It shipped with an upgraded Claude Code and new agent-oriented API features.

The release showed Anthropic's tier strategy evolving — the 'middle' model had become the one most developers actually built on.

OpenAI released GPT-5, the first model to unify its two lineages: a fast model for most queries, a deeper 'thinking' model for hard problems, and a real-time router that picks between them based on the task. Users no longer had to choose a model — the system decided.

OpenAI reported gains over GPT-4 in coding, writing, health answers, and factual accuracy, with lower hallucination rates and stronger agentic tool use. It shipped across ChatGPT, Microsoft Copilot, and the API on day one.

GPT-5 became the base of the 5.x generation, which OpenAI iterated rapidly through 5.1, 5.2, and beyond — a faster cadence than any previous flagship line.

Anthropic released Claude Opus 4.1 as a drop-in replacement for Opus 4, keeping the same pricing and rate limits. The upgrade focused on coding precision, multi-file refactors, and agentic task completion.

Anthropic positioned 4.1 as a refinement rather than a reinvention: better at following complex instructions and producing working code on the first pass. Benchmarks showed gains on agentic coding and tool-use evaluations.

The release continued Anthropic's mid-generation refresh pattern, squeezing more capability out of the Claude 4 architecture before the next numbered leap.

xAI released Grok-4, which Musk called 'the most intelligent model in the world,' with native tool use and stronger reasoning. Alongside it came Grok-4 Heavy, which runs multi-agent ensembles on hard problems, sold through a $300-per-month SuperGrok Heavy tier.

Grok-4 Heavy's ensemble approach — spawning multiple agents to work a problem in parallel — previewed the multi-agent patterns the whole industry adopted in 2026. The $300 tier was the priciest consumer AI subscription to date.

Grok-4 held xAI's frontier position through the second half of 2025, with the 4.1 refresh arriving in November.

Anthropic launched the Claude 4 generation with two models: the flagship Claude Opus 4 and the workhorse Claude Sonnet 4. Both were pitched as the best coding models in the world, capable of sustaining coherent work over multi-hour agentic sessions.

Opus 4 introduced the ability to use tools in parallel and maintain focus across long-horizon tasks, while Sonnet 4 delivered most of that capability at a lower price. Anthropic framed the pair around agents that could work autonomously for extended periods.

Claude 4 reset the frontier in software engineering tasks and kicked off the rapid 4.x iteration cadence — 4.1, 4.5, 4.6, and 4.8 followed within a year.

Google released Gemini 2.5 Pro as an experimental preview, a state-of-the-art reasoning model that led major benchmarks through the middle of 2025. It combined the 2.x generation's tool use with substantially stronger step-by-step thinking.

The experimental label didn't stop developers from adopting it widely; 2.5 Pro became the model to beat on coding and math evaluations for months. Google iterated it rapidly through the spring before a stable release.

Gemini 2.5 Pro's long benchmark reign set the stage for the Gemini 3 generation, which Google teased as a bigger leap later in 2025.

OpenAI launched GPT-4.5, codenamed Orion, describing it as its largest model to date. It was positioned as a research preview for Pro users and developers — notably, the last OpenAI model released without chain-of-thought reasoning.

GPT-4.5 emphasized emotional intelligence and conversational quality over raw benchmark gains. Reviews found it warmer and more intuitive than the reasoning models, but its huge training cost made it expensive to serve.

In retrospect it marked the end of an era: the final flagship of the pure next-token-prediction lineage before the GPT-5 generation unified fast and reasoning models under one system.