Our public Model Pricing page has one job: show the real, published cost of every model ApiSpi routes to — the same numbers your Token Bank is billed at, with no markup. This month it learned three new tricks.

The GPT Lineup, Priced

The catalog now carries fifteen priced OpenAI models, taken from OpenAI's own pricing documentation: the new GPT-5.6 family (Sol, Terra, and Luna), the GPT-5 tier from nano to Pro, GPT-4.1 and GPT-4o with their minis, and the o-series reasoning models. GPT-5.6 Terra — the balanced mainstream tier — is the billing headline, with rates spanning from GPT-5 nano at five cents per million input tokens to GPT-5 Pro at fifteen dollars.

What a Cache Hit Costs

If you read our guide to prompt caching, you know repeated prompt prefixes can be served from cache at a fraction of the normal input rate. The pricing table now makes that visible: a Cached / 1M column lists each provider's published cache-read rate — OpenAI's cached-input prices, Anthropic's cache reads at one-tenth of base input, and newly published rates for GLM 5.2 and Qwen 3.8 Max on SCX. Where a provider publishes no cache rate, we show a dash rather than a guess.

Seeing the Whole Market at Once

Tables answer "what does this model cost?"; they're poor at "where does everything sit?". The new price-landscape chart plots every priced model by input cost against output cost on logarithmic axes — cheap and cheerful in the bottom-left, premium reasoning in the top-right — coloured by provider, with a hover callout naming each point. The table itself also got friendlier: rounded row cards, column-heading filters with searchable dropdowns, and click-through to each model's detail page, where structured specifications (parameters, context window, licence) now sit alongside the price history chart.

Benchmarks Joined the Party

The Model Benchmarks page picked up the LMArena text leaderboard — the crowd-vote Elo board — alongside its sourced coding, multimodal, and multilingual benchmarks, each now rendered as a ranked bar chart with a cross-benchmark "who leads" summary on top. Between the pricing observatory and the benchmark boards, you can now answer both halves of the model-selection question — what it costs, and what you get for it — without leaving the site.

Why We Bother

Model pricing drifts constantly and vendors rarely make comparison easy. Keeping a public, sourced, always-current record is our answer: it's the page our own routing and billing run from, so it can't quietly go stale. When a rate changes upstream, it changes here — and your Token Bank follows the same number you can see.