AI Benchmark Index
Global Model Rankings
Which AI model is #1 right now?
New #1 · AA Index v4.3
Claude Opus 5.5
58
AA Index
Release Radar
What shipped this month
October so far: Google's Gemini 4 Argon got its first independent score, Anthropic filled out the 5.5 family with Haiku, Mistral went trillion-scale open, and image + voice models refreshed. Scores use AA Index v4.3.
OCT 2026
Google
Gemini 4 Argon
AA 531M ctx$2 / $10 promo
Google's first Gemini 4 model scores 53 on AA v4.3 — tied with GPT-6 Astra, #4 overall, behind Opus 5.5 (58). #1 on AutomationBench (78%) and the lowest hallucination rate among top models (15%). Still limited to Fairwind security partners; paid API + AI Ultra are next.
Anthropic
Claude Haiku 5.5
AA 43$0.10 / $0.50Small tier
Completes the 5.5 family — small, fast model at ~75% cheaper than Haiku 4.5, yet scores 43, above GPT-6 Luna (38). The budget pick for high-volume work.
Mistral
Mistral Large 4
1.05T / 52BOpen1M ctx
Europe's first trillion-parameter open-weight MoE, multimodal. AA 38 at launch — behind GLM-5.3 (45) and Kimi K3 (44). Launch sale $0.68/$2.09; weights promised by end of October.
Google · BFL
Nano Banana 2.1 · FLUX 3 Image
Image4K10 refs
Nano Banana 2.1 halves image-output price vs Nano Banana 2; FLUX 3 Image adds 4K generation + editing with up to ten reference images. Arena scores still pending — GPT Image 2.5 holds #1.
Microsoft
MAI-Voice-2.1 · MAI-Transcribe-2
TTS23 langsSTT #1 WER
New TTS in 23 languages (Flash variant ~150 ms latency) plus a streaming speech-to-text model Microsoft says ranks #1 on AA streaming WER (2.5%).
Anthropic
Claude Sonnet 5.5
AA 56#2 overall< ½ Opus price
Mid-tier model lands at #2 on the AA Index — ahead of GPT-6 Astra and Argon — just two points behind Opus 5.5. The default pick for most teams.
Anthropic
Claude Opus 5.5
AA 5820% cheaper1M ctx
Still the overall #1 on AA Index v4.3; leads 6 of 10 evals incl. Humanity's Last Exam 61.4% and GDPval-AA. Shipped with a 20% price cut.
OpenAI
GPT-6 Sol / Luna · GPT-6.1 Sol
AA 52$2 / $10½ cost
GPT-6 Sol and Luna halve cost vs GPT-5.6; GPT-6.1 Sol lifts Sol to ~52 on the index at a fraction of Opus's cost per task.