GPT-6 Blog
Opinionated, practical guides to GPT-6 API pricing — with worked numbers, not marketing copy.
GPT-6 Cached Input Pricing Explained: What a Cache Hit Actually Costs
A cached token costs $0.01–$1.00 per 1M — 5–10% of fresh input. Same request on Sol 6.1: $0.122 uncached vs $0.027 cached, and the write costs plus the 30-minute vs 12-hour choice decide whether the discount lands.
Read article →
Claude Fable 5.2 vs GPT-6 Astra: Anthropic's Flagship Now Costs 20% Less
Anthropic cut Fable to $8/$40 on Nov 9 — the five-week tie with Astra at $10/$50 is over. A heavy request costs $3.60 vs $4.50, and past 272K tokens the gap blows out to 56%.
Read article →
Gemini 2.5 Reroutes on December 1: What Happens to Your Bill
Pro requests silently move to 3.8 Flash: a 10M/2M workload falls from $32.50 to $15.00 in December, then climbs to $30.00 in January. Flash bills don't move — unless token counts do.
Read article →
GPT-6.1 Terra API Pricing: $1/$5 — Half of Sol61, a Tenth of Astra
OpenAI launched Terra on Oct 26 at exactly half of Sol61's rates on every line. One 50K-input turn costs $0.037; 100K requests a month drops from $7,400 to $3,700.
Read article →
How Much Does 1 Million GPT-6 Tokens Cost? $0.22 to $22.00
A realistic mixed million (700K in / 300K out) costs $0.22 on Luna, $4.40 on 6.1 Sol, $22.00 on Astra — and output is 5× input on every tier. The full bill math, drawn to scale.
Read article →
GPT-4o Deprecated: Migrating to GPT-6 Luna Cuts Your Bill ~95%
OpenAI deprecated the GPT-4 series on Oct 14 — shutdown Jan 14, 2027, Luna is the official replacement. Worked math: $325/mo becomes $15/mo, and the Oct 23 o1/o3 shutdown is 7 days out.
Read article →
Gemini 4 Argon API Pricing: $2/$10 Until Nov 30, Then It Doubles
Google's new flagship rents at GPT-6.1 Sol prices for 47 days — then jumps to $4/$20. Same coding task costs $0.40 now, $0.80 in December, $2.00 on Astra. Full rate card and fine print inside.
Read article →
OpenAI Decisions API Pricing: $0.10/M In, Zero Output Fees
OpenAI's GA'd /v1/decisions endpoint bills input at Luna's $0.10/M rate — output free, cache reads discounted. A 1M-request moderation pipeline drops from $100 to $40/month.
GPT-6.1 Sol Ultrafast Pricing: $12/$60 per 1M Tokens — When 6× Pays
OpenAI shipped Ultrafast for GPT-6.1 Sol on Oct 8 — 6× Standard on every line. At $60 per 1M output, the fastest Sol costs more than standard Astra. Bill math inside.
Read article →
Claude Haiku 5.5 vs GPT-6 Luna: Same $0.10/$0.50, Different Breakpoints
Anthropic matched Luna's sticker on Oct 7 — $0.10/$0.50 under 100K tokens. Past 100K Haiku jumps 5× while Luna holds to 272K. Where the tie breaks, in bill math.
Read article →
GPT-6 Astra Ultrafast Pricing: 6× Standard, $300 per 1M Output
Same Astra model, fastest lane OpenAI sells — $60/$300 per 1M tokens. What 6× does to a real monthly bill, and the one comparison that recalibrated it for me.
Read article →
GPT-6.1 Sol API Pricing: Same $2/$10 as Sol, Cached Input Halved
OpenAI launched GPT-6.1 Sol yesterday with headline rates identical to Sol — the savings are all in cached input at $0.10/M. Exact bill math vs Sol, Astra, and Claude Sonnet 5.5.
Read article →
Anthropic Now Bills Refused Claude API Requests
A refused 100K-token request on Opus 5.5 costs $0.40 — for zero output. Which categories bill you, who gets hurt most, and how to budget for it.
Read article →
How Much Does It Cost to Run a Coding Agent on GPT-6?
Light $21.48/mo, Medium $145.20/mo, Heavy $514.80/mo on GPT-6 Sol — illustrative scenarios, not averages. Given workload X, the API cost is Y.
Read article →
10 Ways to Cut Your GPT-6 API Bill
Ten levers that actually move your GPT-6 bill — tier right-sizing, caching, output caps — learned the hard way, with real numbers and zero fluff.
Read article →
GPT-6 Astra vs Sol vs Luna: Which Tier Should You Use?
A 100× price spread between the cheapest and priciest tier. An opinionated, index-card-sized framework for landing on the right rung — with the math drawn to scale.
Read article →
How to Estimate Your GPT-6 API Costs Before You Ship
The formula is one line; the gotchas are the article. Reasoning tokens, retries, history growth — the four things that eat naive cost estimates alive.
Read article →