Frontier models · Updated Oct 8, 2026
The best AI models of 2026, compared.
Release dates, context windows, open or closed weights and list prices for the 23 models the race is being run on, from GPT-6 Astra and Claude Fable 5.1 to Gemini 4 Argon, Grok 4.7, Kimi K3 and DeepSeek V4. Sorted newest first.
23 models
| Model | Released | Tier | Context | Weights | Price (in / out per 1M tokens) | Notes |
|---|---|---|---|---|---|---|
| Gemini 4 Argon Google DeepMind | Sep 30, 2026 | Frontier | Undisclosed input, 1M output | Closed | $2 / $10 intro, then $4 / $20 | Announced after months of delays; limited to cyber-defense partners first. Leads 13 of 19 Google-reported benchmarks. |
| GPT-6.1 Sol OpenAI | Sep 29, 2026 | Frontier | 1M | Closed | $2 / $10 | Near-Astra performance on coding and professional work; multi-agent delegation in beta. |
| Claude Sonnet 5.5 Anthropic | Sep 28, 2026 | Workhorse | 1M | Closed | $2 / $10 | Best combination of speed and intelligence in the Claude lineup. |
| Claude Opus 5.5 Anthropic | Sep 22, 2026 | Frontier | 1M | Closed | $4 / $20 | Recommended starting point for complex coding and long-running agents. Artificial Analysis Index 58. |
| GPT-6 Sol / Luna OpenAI | Sep 22, 2026 | Workhorse | 1M | Closed | $2 / $10 (Sol), $0.10 / $0.50 (Luna) | Agent work (Sol) and high-volume tasks (Luna). |
| Grok 4.7 xAI | Sep 21, 2026 | Frontier | 500K | Closed | $2 / $6 | Text and image input; built for coding, agentic tasks and knowledge work. |
| DeepSeek V4.1 Flash DeepSeek | Sep 10, 2026 | Budget | 1M | Open | $0.30 / $1.20 peak | Image input added; half price off-peak. |
| GPT-6 Astra OpenAI | Sep 3, 2026 | Frontier | 1.05M | Closed | $10 / $50 | OpenAI's most capable model for the hardest reasoning, coding, computer-use and research work. |
| Gemini 3.8 Flash Google DeepMind | Sep 2, 2026 | Workhorse | 1M | Closed | $0.75 / $3.75 (promo) | Fast multimodal model for long-horizon software engineering and agents. |
| Claude Fable 5.1 Anthropic | Sep 1, 2026 | Frontier | 1M (128K output) | Closed | $10 / $50 | Frontier tier for difficult autonomous work; a Mythos-class model. |
| Qwen3.8-Flash Alibaba Qwen | Aug 26, 2026 | Budget | 1M | Open | $0.15 / $0.47 | Open weights; text, image and video input. |
| GLM-5.3 Z.ai (Zhipu) | Aug 18, 2026 | Workhorse | 1M | Open | $1.40 / $4.40 | Long-horizon engineering model from Beijing's Zhipu. |
| DeepSeek V4 Pro DeepSeek | Aug 13, 2026 | Frontier | 1M | Open | $1.32 / $3.96 peak | Low-cost reasoning and coding at roughly a tenth of US frontier prices. |
| Grok 4.6 xAI | Aug 12, 2026 | Workhorse | 500K | Closed | $2 / $6 | Coding, engineering and tool use. |
| Qwen3.8-Max Alibaba Qwen | Aug 3, 2026 | Frontier | 1M | Closed | $2 / $6 | Premium tier; multimodal agents and coding. |
| Claude Opus 5 Anthropic | Jul 31, 2026 | Frontier | 1M | Closed | $5 / $25 | Hard coding and long-running agents; superseded by Opus 5.5. |
| Kimi K3 Moonshot AI | Jul 16, 2026 | Frontier | 1M | Closed | $3 / $15 (gateway) | 2.8T total parameters; frontier long-context reasoning. |
| GPT-5.6 Sol / Terra / Luna OpenAI | Jul 9, 2026 | Workhorse | 1M | Closed | $4 / $20 to $0.20 / $1.20 | Value tiers below the GPT-6 generation. |
| Kimi K2.7 Code Moonshot AI | Jun 15, 2026 | Workhorse | 262K | Open | $0.95 / $4 | Open-weight coding model for long agent trajectories; multimodal input. |
| MiniMax M3 MiniMax | Jun 1, 2026 | Workhorse | 1M | Open | $0.30 / $1.20 | Open weight; natively multimodal. |
| Mistral Medium 3.5 Mistral AI | Apr 15, 2026 | Workhorse | 256K | Closed | $1.50 / $7.50 | Mistral's general frontier tier. |
| Mistral Small 4 Mistral AI | Mar 10, 2026 | Budget | 256K | Open | $0.15 / $0.60 | Apache 2.0; multimodal. |
| Llama 4 Scout / Maverick Meta | Apr 5, 2025 | Workhorse | 10M (Scout) | Open | Host-dependent | Open weights; Behemoth teacher model still unreleased. |
Prices are the provider's direct list price and change often; always check the current rate before relying on it. Tiers are our editorial grouping.