All three OrcaCyber tiers sit behind one OrcaRouter key, so switching between them mid-pipeline costs a model string rather than an integration. The OrcaCyber Zero 1.5 pricing line is $3.00 in and $7.50 out per million tokens, and it sits between two other prices in the same family: $0.80 and $1.60 for the Flash tier, $5.00 and $10.00 for the Ultra flagship [OURS]. Three cards, one line of tuning, and a spread of 6.25× between the cheapest input rate and the most expensive. Choosing between them is arithmetic, not taste, and the arithmetic has a shape worth spelling out before you commit a workload.
The ladder
| Flash 1.0 | Zero 1.5 | Ultra 1.0 | |
| Input / M | $0.80 | $3.00 | $5.00 |
| Output / M | $1.60 | $7.50 | $10.00 |
| Cache / M | $0.15 | $0.75 | $0.75 |
| Context | 1,048,576 | 1,000,000 | 1,048,576 |
| Output ceiling | 393,216 | 128,000 | 131,072 |
| Throughput (7d) | 138.8 tok/s | 114.5 tok/s | 42.2 tok/s |
| Image input | yes | no | no |
Two rows are doing the actual work: output ceiling and throughput. Price tells you what a token costs; those two tell you how many tokens a request will need and how long it will take to get them.
The same request on all three
Take a realistic audit call: 200,000 input tokens of repository and notes, and a 20,000-token report.
- Flash 1.0: 0.2 × $0.80 + 0.02 × $1.60 = $0.192
- Zero 1.5: 0.2 × $3.00 + 0.02 × $7.50 = $0.750
- Ultra 1.0: 0.2 × $5.00 + 0.02 × $10.00 = $1.200
Zero costs 3.9× what Flash costs for the same call, and Ultra costs 6.25×. Nothing in that ordering is surprising; the multiples match the price table.
The surprise is in time. At the measured seven-day throughput, generating 20,000 output tokens takes 144 seconds on Flash, 175 seconds on Zero, and 474 seconds on Ultra [OURS]. If a person is waiting, the difference between two and a half minutes and eight minutes is the difference between a tool and a batch job. If a queue is waiting, it is the difference between one worker and three.
The ceiling changes the comparison
Per-million pricing compares models that produce the same number of tokens. They do not.
Flash can emit 393,216 tokens in one response. Zero stops at 128,000; Ultra at 131,072. A task that genuinely needs a 200,000-token artifact — a full report, a long patch set, a complete transcript rewrite — cannot be done on Zero at any price. It takes two calls, and the second call has to carry the first as context, which changes the token count and often the quality.
Cost at the ceiling is the honest way to compare:
- Flash: 393,216 × $1.60 / 1e6 = $0.63
- Zero: 128,000 × $7.50 / 1e6 = $0.96
- Ultra: 131,072 × $10.00 / 1e6 = $1.31
At maximum output, Flash is cheaper than Zero by a third — despite Zero’s lower absolute ceiling. The per-token comparison reverses depending on how much of the ceiling you use.
What a working month looks like
Per-call prices are easy to argue about and hard to feel. A monthly model at plausible volume makes the difference concrete.
Assume a security team running twenty audits a day, six hundred a month, each audit the 200,000-in / 20,000-out shape from above.
| Tier | Cost per audit | Monthly cost | Generation hours |
| Flash 1.0 | $0.192 | $115 | 24.0 |
| Zero 1.5 | $0.750 | $450 | 29.1 |
| Ultra 1.0 | $1.200 | $720 | 79.0 |
All three are arithmetic on the published rates; the generation hours divide total output tokens by the measured throughput [OURS].
Three observations that do not survive a per-token comparison but show up here.
The gap is smaller than it looks. $115 against $450 is a factor of four, and $450 against $720 is a factor of one and a half. On a security team’s budget, the distance between the middle and top tiers is a few hundred dollars a month — less than the cost of the analyst time one extra failed audit consumes.
The generation hours are the binding constraint. Twenty audits a day on the flagship needs seventy-nine hours of model time a month, which is fine in parallel and impossible in a single worker with a fixed timeout. The cost number says “affordable”; the time number says “provision accordingly”.
The cache ratio decides more than the tier. These figures assume no cache hits. At a ninety percent hit rate — 0.02M fresh input and 0.18M cached per audit — the numbers become $0.075 on Flash, $0.345 on Zero and $0.435 on Ultra per audit [OURS]. Every tier drops by more than half, and the ranking is unchanged: Flash remains roughly four and a half times cheaper than Zero.
The monthly view also exposes a pattern worth naming: the tiers are priced like a ladder, but the workloads that justify each rung are not evenly spaced. Most teams need the bottom rung and occasionally escalate a single request. Few need to run the whole month on the top.
The cache line is where the tiers diverge most
Cache reads are priced at 25% of input on Zero 1.5, 15% on Ultra, and 18.75% on Flash [OURS]. All three are expensive relative to the cheapest card in the wider catalogue, where a cache read is two percent of input — but within this family the ordering is Zero worst, Ultra best, Flash in between.
For an agent loop that carries a 100,000-token system prompt and hits cache on most calls, the effective price is not the headline rate. On Zero, a call that is ninety percent cache costs 0.1 × $3.00 + 0.9 × $0.75 = $0.975 per million input tokens — a third of the headline. On Flash the same call costs 0.1 × $0.80 + 0.9 × $0.15 = $0.215. The gap widens from 3.75× on headline rates to 4.5× here — a cache-heavy loop makes Zero relatively more expensive, not less.
Picking a tier
All three cards carry the same access gate — an engagement, a passkey and acceptance of the terms — so there is no cheaper on-ramp at the bottom of the ladder; the gate is a family-level condition, not a per-card one [OURS].
Flash 1.0 for volume, for anything that needs image input, and for anything that needs more than 128,000 output tokens. It is the cheapest card in the family and the fastest.
Zero 1.5 for security reasoning where the refusal-free behaviour and the published profile matter more than the per-token rate, and where the artifact fits in 128,000 tokens. Access is gated to the Security Research tier [OURS].
Ultra 1.0 only when the extra reasoning depth justifies a 42-token-per-second card. At $5 and $10 it is 6.25× the Flash input rate for the same million tokens, and the page publishes no benchmark rows to quantify the difference. That is a decision you make on evidence from your own workload, not from a table.
The takeaway
The ladder is 6.25× from bottom to top, and throughput falls 138.8 → 114.5 → 42.2 tokens per second as the price rises. For an audit-shaped call the per-request cost runs $0.19, $0.75 and $1.20 respectively; at the output ceiling the ordering flips between the bottom two.
Pick the tier by the shape of the artifact and the patience of the consumer, not by the headline rate.
Sourcing note: all rates, context windows, output ceilings and throughput figures are OrcaRouter catalogue data and production telemetry for the seven days ending 2026-10-26 [OURS]. Per-request costs and ceiling costs are arithmetic on those rates. No benchmark scores are used in this article. Access gating for Zero 1.5 is stated on the model page.