Model release
Every major model launch, rewritten as a cost-and-switch analysis. The prices and scores on each page are pulled live from the provider pages, not copied into the article.
MiniMax M3.1 Flash costs $0.15 per 1M input and $0.60 per 1M output — exactly half of MiniMax M3, with a flat rate and a 1,000,000-token total window. Three comparable scores, all vendor-measured.
DeepSeek V4.1-Flash 1017 (2026-10-17) keeps the $0.30 / $1.20 peak price and the $0.006 cache read, and publishes 11 new benchmark scores. Here is what changed, what it costs, and whether you should switch.
Google's Gemini 4 Argon costs $2 per 1M input and $10 per 1M output until 2026-11-30, then $4 / $20 — exactly what Claude Opus 5.5 costs today. A 1,048,576-token window with no long-context step, and ten verified scores out of Google's own nineteen-row evaluation table.
Claude Opus 5.5 costs $4 per 1M input and $20 per 1M output — 20% below the retired Claude Opus 5 — and reads cached input at 5% of the input rate. Four vendor-reported scores, a 1M-token context window, and a September 22 launch.
Grok 4.7 costs $2 per 1M input and $6 per 1M output, reads cached input at $0.50, and doubles all three rates once a prompt reaches 200K tokens. The window is 500,000 tokens; the score card is two benchmarks.
DeepSeek V4.1-Flash 0910 — the 2026-09-10 launch checkpoint — costs $0.30 per 1M input and $1.20 per 1M output at peak, $0.15 / $0.60 off-peak, and reads cached input at $0.006, with 19 verified benchmark scores. The API now serves the newer 1017 checkpoint, tracked separately.
GPT-6 astra costs $10 per 1M input and $50 per 1M output — 5× GPT-6 sol, 100× GPT-6 luna, and still cheaper than GPT-5.6 cyber. Four verified scores, a 1,050,000-token window, and long-context pricing above 272K input.
Google's Gemini 3.1 Pro Preview costs $2 per 1M input and $12 per 1M output up to 200,000 tokens, $4 / $18 above that, inside a 1,000,000-token window. Five verified benchmark scores and a long-context tier the older page missed.