LLM API prices show wide gap across providers

Pricing verified June 28, 2026 shows a wide gap among hosted LLM APIs. DeepSeek V4 Flash is listed as the cheapest option at $0.14/$0.28 per 1M tokens, while GPT-5.5 is priced at $5/$30 and Claude Opus 4.8 at $5/$25. MiniMax M3 at $0.60/$2.40 is framed as the strongest value pick, with a score above 80% on SWE-bench Verified and a 1M context.

Quality rankings are concentrated near the top for coding tasks. GPT-5.5 scores 88.7% on SWE-bench Verified and Claude Opus 4.8 scores 88.6%, while Anthropic’s higher-scoring Fable 5 at 95.0% was suspended in June 2026. Several lower-cost models, including DeepSeek V4 Pro Max, Gemini 3.1 Pro, MiniMax M3, Qwen3.7 Max, and Kimi K2.6, fall between 80.2% and 80.6%.

Operational differences are as important as headline rates. Anthropic uses spend-based tiers and cache-aware limits, OpenAI unlocks tiers by payment history, and DeepSeek publishes concurrency caps of 2,500 on V4 Flash and 500 on V4 Pro. Most providers support OpenAI-format requests, and the guidance favors routing workloads across cheap, frontier, and specialized APIs rather than standardizing on one model.