AI Models
Guide

The versatile all-rounder with dynamic internal routing.

• GPT-5.5 is the current flagship (Apr 23, 2026) — $5/1M input, $30/1M output; 1M token context; improved coding, computer use, and multi-step research
• GPT-5.5 rolling out to Plus, Pro, Business, and Enterprise via ChatGPT and Codex
• Uses an internal router to select the right sub-model per request in real time
• GPT-5.4 Thinking (Mar 5, 2026): reasoning-first variant — strongest at math, spreadsheets, research, and document tasks
• GPT-5.3-Codex (Feb 5, 2026): dedicated agentic coding variant combining Codex and GPT-5 training stacks
• GPT-5.3 Instant: fast chat variant for low-latency applications
• Native computer use and tool calling for agentic automation

The developer favorite for coding, reasoning, and safety.

• Opus 4.7 (Apr 15, 2026): latest GA model — 1M context window, $5/1M input, $25/1M output; step-change agentic coding, substantially improved vision resolution
• Sonnet 4.6: best speed/cost balance for most production workloads
• Haiku 4.5: fast and cheap for high-volume tasks
• Leads human-preference leaderboards; strong ARC-AGI-2 scores
• Agentic capabilities: autonomous multi-step coding and long-horizon tasks
• Claude Mythos Preview (Apr 6–7, 2026): invite-only cybersecurity research model via Project Glasswing — limited to ~12 enterprise partners; not publicly available

Multimodal powerhouse with top benchmark breadth.

• Gemini 3.1 Pro leads the Artificial Analysis Intelligence Index
• Native multimodal: processes text, images, audio, and video natively
• 1M token context window; deep Google Workspace integration
• Gemini 3 Flash: current default model in the Gemini app — strong balance of speed and capability
• Gemini 3.1 Flash-Lite (Mar 3): $0.25/1M input, 2.5× faster than prior Flash — cost-efficient tier for high-volume workloads
• Gemini 3.1 Flash Live: audio/voice model with 90+ language support and lower latency (March 26)
• Gemini 3.1 Flash TTS: native text-to-speech model with audio generation capabilities

Real-time data meets raw reasoning power.

• Grok 4.20 uses native multi-agent architecture with four specialist agents that debate before responding
• Grok 4.20 Beta 2 (March 3): improved instruction following and hallucination reduction
• Real-time integration with X (Twitter) for current events
• Grok 4 Heavy (extended inference variant) scored 100% on AIME 2025 — not the base Grok 4.20 model; attribution is widely misreported
• Grok 5 in training — reported ~6T parameter MoE model training on xAI's Colossus 2 supercluster; Q2 2026 target

Search-native AI built for grounded, cited answers.

• Built on Llama 3.3 70B, further trained for search-grounded factuality
• Runs at 1,200 tokens/sec on Cerebras inference hardware
• Model family: Sonar, Sonar Pro, Reasoning Pro, Deep Research
• Matches GPT-4o on user satisfaction benchmarks

Frontier-level coding model trained via RL on long-horizon software engineering tasks.

• 73.7 on SWE-bench Multilingual — top scores across CursorBench and Terminal-Bench 2.0
• Trained with RL to solve complex tasks requiring hundreds of sequential actions
• Fast variant is now default: frontier-level intelligence at lower cost than competing fast models
• Background agents run tasks autonomously while you work

Microsoft's own foundation model stack — independent of OpenAI, built for speech, voice, and vision.

• MAI-Transcribe-1: speech-to-text across 25 languages; outperforms Whisper-large-v3 on accuracy
• MAI-Voice-1: generates 60s of audio in 1s; supports voice cloning
• MAI-Image-2: high-quality image generation
• All available on Microsoft Foundry — signals Microsoft building foundational AI independent of OpenAI (Apr 2, 2026)

Fine-tuned GPT-5.4 for enterprise defensive cybersecurity — gated behind OpenAI's Trusted Access for Cyber program.

• Released April 14, 2026 — fine-tune of GPT-5.4 for dual-use security research
• Lowered refusal thresholds for defensive cybersecurity tasks; native binary reverse engineering without source code
• Gated behind OpenAI's Trusted Access for Cyber (TAC) program; enterprise partners include CrowdStrike, Cloudflare, Palo Alto Networks, Cisco, JPMorgan, Goldman Sachs
• No public API pricing; $10M in API credits committed via Cybersecurity Grant Program
• Direct counterpart to Anthropic's Claude Mythos Preview — part of a matched pair of restricted cyber-focused models released within days of each other

Invite-only cybersecurity research model via Project Glasswing — Anthropic's counterpart to GPT-5.4-Cyber.

• Released April 6–7, 2026 via Project Glasswing — not publicly available
• GPQA 0.9; 93.9% SWE-bench Verified; 97.6% USAMO 2026
• Limited to ~12 enterprise partners: AWS, Apple, Cisco, CrowdStrike, Google, JPMorgan, Microsoft, NVIDIA, and others
• Priced at ~$2,500/1M tokens to gate general use and prevent misuse
• Cybersecurity-focused capabilities not available in standard Claude models

The leading open-source model family.

• Llama 4 Scout: industry-leading 10M token context window
• Llama 4 Maverick: 17B active / 128 experts — outperforms GPT-4o and Gemini 2.0 Flash on key benchmarks
• Fully open weights; can be self-hosted for complete data control
• Llama 4 Behemoth (288B active) still in training and unreleased as of April 2026; widely covered as next milestone

Cost-redefining open-source frontier — released April 24, 2026 in two variants.

• V4-Pro: 1.6T total parameters / 49B active MoE; $1.74/1M input, $3.48/1M output — cheapest frontier-class open model
• V4-Flash: 284B total / 13B active; $0.14/1M input, $0.28/1M output — ultra-budget tier
• Apache 2.0 license; native 1M context window; built-in agentic long-context and tool-use
• Engram conditional memory and Manifold-Constrained Hyper-Connections for improved long-context performance
• Entire V3/V4 lineage trained for under $6M — redefining AI cost efficiency

The enterprise-safe European model family, now spanning text, reasoning, code, and speech.

• Mistral Large 3: EU AI Act-compliant flagship for regulated industries (finance, healthcare, gov); 675B total MoE
• Magistral: Mistral's reasoning model — multilingual, transparent chain-of-thought
• Mistral Small 4 (March 16): 119B/6.5B-active MoE unifying reasoning (Magistral), vision (Pixtral), and coding (Devstral) in one endpoint
• Devstral 2: 72.2% SWE-bench Verified — top open agentic coding model; 123B params, 256K context, MIT license; competitive with Composer 2
• Voxtral (March 26): open-source 4B text-to-speech, 9 languages, runs on consumer hardware
• Strong European data sovereignty guarantees across the full model family

Open-source agentic leader — top SWE-Bench Pro score and the largest open agent swarm.

• Released April 20, 2026 — 1T parameter MoE, 32B active, 262.1K context window
• Agent Swarm scales to 300 specialized sub-agents with up to 4,000 coordinated steps (was 100)
• 58.6 on SWE-Bench Pro — edges GPT-5.4's 57.7; top open model on this benchmark
• Open-weight, Modified MIT license; weights on Hugging Face
• Kimi Code CLI agent rivals Claude Code and Gemini CLI

Frontier-class model on a MIT license.

• Updated Apr 6, 2026 — improved from GLM-5 baseline
• 744B parameter MoE model (44B active) with 200K context window
• Released under MIT license; trained entirely on Huawei Ascend chips (zero NVIDIA GPUs)
• GPQA 0.9; 77.8% on SWE-bench Verified; 50.4% on Humanity's Last Exam
• Priced roughly 6x cheaper than comparable proprietary models

NVIDIA's open agentic reasoning stack — Nano, Super, and Ultra sizes on Bedrock.

• Released March 2026; available on Amazon Bedrock and NVIDIA NIM
• Three sizes: Nano (edge/on-device), Super (balanced), Ultra (frontier-class agentic reasoning)
• Nemotron 3 Super peers with Llama 4 Maverick on open-model benchmarks
• Leading open model for agentic reasoning and multi-step tool use
• Optimized for NVIDIA hardware; available for self-hosting via NIM microservices

Open-weight models from Gemini 3 research — optimized for on-device and frontier-class performance.

• Four Apache 2.0 models: E2B (2.3B), E4B (4.5B), 26B MoE (4B active), 31B dense (Apr 2, 2026)
• 31B ranks #3 on Arena AI leaderboard at 1452 Elo — outperforms models 20× its size
• E2B/E4B optimized for on-device Android: up to 4× faster and 60% less battery than prior Gemma
• All models natively multimodal; larger variants support 256K context

The multilingual giant — open-weight Qwen3.5 and closed-source agentic Qwen3.6 Plus.

• Qwen3.5 (Feb 2026): open-weight, 397B parameters — available for self-hosting and fine-tuning
• Qwen3.5-Omni: native audio/video/text multimodal — Thinker architecture, 256K context, 113-language speech recognition
• Qwen3.6-Plus (Apr 2, 2026): closed-source API-only, 1M context; agentic — matches Claude Opus 4.5 on SWE-bench and Terminal-Bench 2.0
• 0.6B to 235B open-weight range; Qwen3-Max (1T+) API-only; supports 119 languages
• Qwen3-Coder achieves 69.6% on SWE-Bench Verified, surpassing many frontier models