Anthropic's fastest and cheapest current-gen model. Optimized for high-throughput, low-latency tasks.

New free/Pro default on claude.ai; near-Opus quality at a fraction of the cost.

Top Anthropic SWE-bench Verified score, favorite inside Claude Code/Cursor; uses adaptive-thinking effort parameter instead of extended-thinking budgets.

Mythos-tier flagship; tops SWE-Bench Pro at 80.3%, purpose-built for long-horizon autonomous coding runs. Briefly offline Jun 12–Jul 1 due to export controls.

74.9% SWE-bench, 92.8% GPQA; cheaper mainstream tier alongside mini/nano variants.

Leads FrontierMath Tier 4 at 39.6% — best for hardest math and physics problems.

Second-highest on Artificial Analysis Intelligence Index; strong all-rounder with Canvas editor for writing.

Cheapest tier — edge and small-task models. Gemma 3n from Google, Nova Micro from Amazon Bedrock.

Flash-Lite is the cheapest mainstream option at $0.10/M input. Covers Pro, Flash, and Flash-Lite variants.

Budget-friendly mid-tier option from Google's Gemini 3 generation.

1,656 GDPval-AA Elo — price-performance pick for bulk content and agent-style coding.

Leads reasoning benchmarks; strong video and audio multimodality.

Leads raw SWE-bench at 75%. Purpose-built for software engineering tasks.

xAI's most intelligent and fastest Grok; reasoning, function calling, native web/X search, remote MCP support.

284B total, 13B active params; extremely cheap, MIT licensed open-weights model.

1.6T params, 49B active MoE, MIT license, open weights; 80.6% SWE-bench Verified — highest open-weights score.

99.1% AIME, 84.5% GPQA — best quality-per-dollar in its class.

97.1 on HMMT Feb 2026, best for competition math at half GPT-5.5 Pro's cost.

Largest ecosystem of fine-tunes and community tooling. Fully open weights under Meta's community license.

European option with strong multilingual support; licensing varies by model variant.