Ultramarine 1 Flash Pricing

Ultramarine 1 Flash costs $0.15 per million uncached input tokens, $0.03 per million cached input tokens, and $0.50 per million output tokens on the official Paper Instruments application programming interface (API). Paper Instruments announced the model on October 14, 2026, pointing it at professional knowledge work across finance, law, consulting, and operations.

Every request bills output tokens for visible response text and internal reasoning combined, drawing down from an account balance funded through Stripe Checkout. As of October 16, 2026, Paper Instruments publishes no free usage tier, and teams must add prepaid credits before sending live API requests.

Ultramarine 1 Flash Price per Token

The official Paper Instruments price card lists Ultramarine 1 Flash at $0.15 per million uncached input tokens, $0.03 per million cached input tokens, and $0.50 per million output tokens. These rates were verified on October 16, 2026, through the public pricing endpoint that Paper Instruments serves without authentication.

Output token billing includes both the final response text and intermediate reasoning steps generated during execution. Ultramarine 1 Flash streams its reasoning in the delta.reasoning_content field. Every one of those reasoning tokens bills at the $0.50 output rate, so long reasoning runs raise the cost of a request. The model operates under the model identifier ultramarine-1-flash on the Paper Instruments developer platform.

Cached tokens receive an 80 percent discount relative to uncached input tokens. When an application passes recurring context across multiple turns, Paper Instruments charges $0.03 per million tokens for prompt cache hits. Third-party model router OrcaRouter reports that these published token rates match the base pricing of GLM-5.3-Flash, the open base model that Paper Instruments post-trained.

• Uncached input: $0.15 per million tokens ($0.00015 per thousand tokens)
• Cached input: $0.03 per million tokens ($0.00003 per thousand tokens)
• Output and reasoning: $0.50 per million tokens ($0.0005 per thousand tokens)
• Public rate verification: GET https://api.paperinstruments.com/v1/pricing requires no API key

Billing Mechanics and Account Funding

Paper Instruments handles Ultramarine 1 Flash billing through prepaid balances managed with Stripe Checkout. Developers buy credits in advance through Stripe Checkout before the API processes requests. Paper Instruments credits the balance after the Stripe payment verifies.

The billing documentation describes an optional automatic top-up that charges a saved card. Invoicing terms and post-paid enterprise billing are not listed in the standard developer documentation. As of October 16, 2026, Paper Instruments publishes no free tier, trial credit allocation, or promotional grants for newly registered accounts.

Usage limits apply at the organization level rather than to individual API keys. Developers can read their organization's rate limits by sending an authenticated request to GET /v1/usage. The rate limits reference notes that teams requiring higher request throughput must contact Paper Instruments support directly.

• Prepaid account balance deposited through Stripe Checkout
• Optional automatic top-up from a saved card
• No published free tier or complimentary startup credits on account creation
• Organization rate limits readable via GET /v1/usage

Token Rate Comparisons Across Frontier Models

Ultramarine 1 Flash carries lower per-token card rates than competing frontier models such as Claude Fable 5. Anthropic lists Claude Fable 5 pricing at $10.00 per million input tokens, $1.00 per million cached input tokens, and $50.00 per million output tokens. Anthropic also offers Claude Fable 5.1, which launched on September 1, 2026, maintaining the same $10.00 and $50.00 rates while reducing cached input reads to $0.25 per million tokens.

Comparing published card rates shows wide per-token price ratios between the two providers. At $10.00 versus $0.15 per million tokens, Claude Fable 5 uncached input costs 66.67 times as much as Ultramarine 1 Flash uncached input. At $50.00 versus $0.50 per million tokens, Claude Fable 5 output costs exactly 100 times as much as Ultramarine 1 Flash output.

These calculations describe token sticker prices rather than total invoice costs. Because different models generate varying reasoning volumes and require distinct prompt sizes to complete a workflow, per-token differences do not map directly to final task expenses. The Ultramarine 1 Flash vs Claude Fable 5 comparison puts benchmark scores and reasoning behavior side by side.

Benchmark Task Costs Versus Per-Token Rates

Paper Instruments reports an average cost of $0.13 per task for Ultramarine 1 Flash on its internal PI Knowledge Work benchmark suite. In the company's technical report, Paper Instruments evaluates Ultramarine 1 Flash against commercial alternatives using official October 2026 API prices. On this internal benchmark, Paper Instruments reports a cost of $8.19 per completed task for Claude Fable 5.

Paper Instruments CEO Daanish Khazi cited this benchmark comparison in an announcement thread on X to support the claim that the model is sixty times cheaper than Claude Fable 5. Dividing the reported $8.19 task cost by $0.13 yields approximately 63 times, which matches that claim. This sixty-fold figure comes only from PI Knowledge Work, Paper Instruments' internal suite, which is not public. It is a per-task ratio, and the per-token ratios differ.

The technical report records per-task costs for competing models on PI Knowledge Work: Claude Opus 5.5 reached $6.96, OpenAI GPT-5.6 Sol reached $2.09, and OpenAI GPT-6 Sol reached $1.26 per task. Paper Instruments states that its reinforcement learning (RL) post-training produced 6.5 percent fewer output tokens than the base GLM-5.3 Flash model across six knowledge benchmarks while improving performance. As of October 16, 2026, no outside testing lab has reproduced these task cost figures.

• Ultramarine 1 Flash: $0.13 per task (score 59.3 on PI Knowledge Work, reported by Paper Instruments)
• Claude Fable 5: $8.19 per task (score 57.8, reported by Paper Instruments)
• Claude Opus 5.5: $6.96 per task (reported by Paper Instruments)
• GPT-5.6 Sol: $2.09 per task (score 55.1, reported by Paper Instruments)
• GPT-6 Sol: $1.26 per task (score 61.3, reported by Paper Instruments)
• Token efficiency: 6.5 percent fewer output tokens than base GLM-5.3 Flash across six benchmarks (reported by Paper Instruments)

Application Settings That Determine Generation Spend

Application developers control Ultramarine 1 Flash expenses primarily through completion token limits and prompt cache hit rates. According to the developer documentation, the reasoning_effort parameter defaults to max, and reasoning cannot be turned off. Because internal reasoning tokens count toward the $0.50 per million output token price, workloads that trigger long reasoning cycles generate higher per-call charges.

The max_completion_tokens parameter controls the combined total for visible text and internal reasoning. If an API call cuts off with a finish_reason of length, developers must raise this ceiling to allow the model to conclude its response.

Cached input is the cheapest token class on the Ultramarine 1 Flash price card. Passing the optional x-session-id header gives sticky routing, so consecutive turns of one conversation go to the same backend. Cached input tokens cost $0.03 per million, 80 percent less than the $0.15 uncached input rate. Output tokens still bill at $0.50 per million.

• Keep reasoning_effort at max as recommended by Paper Instruments, but track output token consumption
• Set max_completion_tokens high enough to prevent truncating reasoning before the final answer renders
• Include the x-session-id header to maintain sticky routing across multi-turn agent execution
• Reuse the same context across requests so more input bills at the $0.03 cached rate

Unpublished Operational Specifications and Hosting Limits

Paper Instruments leaves several operational parameters unpublished, including default rate limits and context window caps for production API calls. The official documentation does not list the maximum context window supported on the public API endpoints. While the training run used 131,072 tokens and one internal evaluation ran at 262,144 tokens, Paper Instruments publishes no official context ceiling for production callers.

Default organization rate limits for requests per minute and tokens per minute are also missing from public documents. The rate limits page provides illustrative examples rather than fixed guarantees, directing callers to inspect active quotas via GET /v1/usage. Maximum output token limits per completion are likewise omitted.

Self-hosting expenses remain unknown because Paper Instruments has not released the model weights as of October 16, 2026. In his initial announcement, CEO Daanish Khazi stated weights would arrive roughly two weeks after launch, pointing toward late October 2026.

Workload Fit and Procurement Criteria

As of October 16, 2026, Ultramarine 1 Flash bills only from a prepaid balance and has no published weights license. Organizations that need post-paid invoicing or a license on file before deployment are a poor fit for now. Teams that need an established weights license should review GLM 5.3 Explained to inspect its terms, or evaluate GLM-5.3-Flash directly. Workloads that depend on independent benchmark reproduction should wait for third-party evaluations before committing infrastructure budgets.

At Layer3Labs, our assessment of the financial tradeoff would change if Paper Instruments publishes transparent default organization rate limits, introduces automated trial credits, or releases weights under an unrestricted commercial license. Independent validation from evaluation platforms like Artificial Analysis would also confirm whether reported task savings persist outside Paper Instruments' internal test harness.

Before routing production traffic, query GET https://api.paperinstruments.com/v1/pricing for current rates and GET /v1/usage for the organization's rate limits, then forecast Ultramarine 1 Flash spend from those numbers.

• Avoid Ultramarine 1 Flash if your procurement process requires monthly invoicing
• Consider GLM-5.3 Flash directly if you need to self-host an open model before Ultramarine weights ship
• Monitor independent evaluation releases to verify task cost claims against commercial alternatives
• Calculate token budgets assuming maximum reasoning depth on every production call