Frontier hosted models and open-weight models you can self-host, side by side. Figures as of July 2026 and refreshed regularly.
Pick any two models — up to four — and see how they stack up across coding, agentic, reasoning, knowledge and context, with the key spec differences called out.
The Overall score auto-refreshes weekly from BenchLM (July 2026). The five radar axes (coding, agentic, reasoning, knowledge + a normalised context score) are a manually-curated BenchLM snapshot as of June 2026; some open-model axis values are indicative pending full per-benchmark data, and frontier parameter counts are undisclosed. Confirm specifics on each model’s official page.
Leading hosted models, accessed via API. Scores and prices move quickly; figures are indicative.
| Model | Lab | Context | Notable | Access |
|---|---|---|---|---|
| Claude Fable 5 | Anthropic | 1M+ | Tops the Artificial Analysis Intelligence Index (~65); first public Mythos-class model (Jun 9). Access intermittent under US export controls. | API |
| Claude Opus 4.8 | Anthropic | 1M+ | Highest-scoring widely-available model (AA Index ~61) | API |
| GPT-5.5 | OpenAI | 1M+ | Strong all-round; Pro / Instant variants (AA Index ~60) | API |
| Gemini 3.1 Pro | 1M+ | Top reasoning & data analysis (94%+ GPQA Diamond) | API | |
| Grok 4.3 | xAI | ~2.0M | Cheapest of the frontier four; strong agentic / tool use | API |
| Gemini 3.5 Flash | 1M | Flagship-level quality at ~4x speed (~$1.50 / 1M in) | API |
Sources: LLM-Stats, Morph, LM Council (July 2026).
Open-weight models you can download and self-host — run on your own GPUs or on Semifly in one click.
| Model | Developer | Params | Context | License | Download | Run |
|---|---|---|---|---|---|---|
| DeepSeek V4 Pro | DeepSeek | ~1.6T (MoE) | 1M | MIT | Hugging Face → | Run on Semifly |
| MiniMax M3 | MiniMax | 428B (23B active) | 1M | Apache 2.0 | Hugging Face → | Run on Semifly |
| GLM-5.1 | Zhipu AI | MoE | 200K | MIT | Hugging Face → | Run on Semifly |
| Kimi K2.7-Code | Moonshot AI | ~1T (A32B, MoE) | 256K | Mod. MIT | Hugging Face → | Run on Semifly |
| Qwen3.5 (397B-A17B) | Alibaba | 397B (17B active) | 256K | Apache 2.0 | Hugging Face → | Run on Semifly |
| Llama 4 Scout | Meta | MoE | 10M | Llama 4 | Hugging Face → | Run on Semifly |
| Llama 4 Maverick | Meta | MoE | 1M | Llama 4 | Hugging Face → | Run on Semifly |
| Mistral Small 4 | Mistral AI | 24B | 256K | Apache 2.0 | Hugging Face → | Run on Semifly |
| Gemma 4 (31B) | 31B | 256K | Gemma | Hugging Face → | Run on Semifly |
Confirm the current license on each model’s official page before deployment.
