BLENDED MODELSEVERAL LLMS PER REQUESTPRIVATE BETAIMAGE INPUTPRICE CUT
Pareto API Pricing
Pareto is Unbiased's blended model: each request goes to several frontier and open models in parallel, and one synthesized answer comes back under the model string pareto. It costs $0.80/M input, $0.03/M cached and $3.20/M output - one bill, whatever runs underneath. Unbiased does not name the models in the blend and says the composition can change. The card moves with the release - up on September 18 with 26.9, down by October 4 after 26.10 Preview became the current release - and API access is a private beta with every signup approved by hand.
Input - per 1M tokens
$0.80/M
Fugu Max $2.00/M, also multi-model -68% vs 26.9
Output - per 1M tokens
$3.20/M
Fugu Max $6.00/M -57% vs 26.9
Cached input - per 1M tokens
$0.03/M
Cache read price -96%
Effective - agentic blend
$0.411/M
92/8 split - 82% cache hit rate
§ 01 / TERMINAL
Run the numbers.
Calculator pre-loaded with Pareto's pay-as-you-go rates. Unbiased bills the tokens of your request and its answer at these rates - one model string, one bill - and publishes no separate charge for the models that run underneath. It publishes no context window either, so the long-input end of any estimate is untested.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Real-world presets.
CODING AGENT
Repo-wide bug fix
$0.052/task
CODE REVIEW
Pull request review
$0.016/review
RESEARCH
Research planning turn
$0.00943/turn
ASSISTANT
Interactive chat turn
$0.00357/turn
Estimate · blended-tokenizer-estimate · ≈3.75 chars/token Auto-counts as you type
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
Characters 282
Words 46
Tokens (estimated) 75 tokens
Cost as input · uncached $0.00006 USD
Cost as output · uncached $0.00024 USD
Cost as cached input $0.000002 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Pareto Current | $0.80 cache $0.03 | $3.20 | $0.411 agentic 92/8 | not published | One API call with several models behind it |
| Fugu Max | $2.00 cache $0.25 | $6.00 | $1.00 +2.4× | 1M | Multi-model orchestration at a fixed card |
| Fugu Ultra | $5.00 cache $0.50 | $30.00 | $3.61 +8.8× | 1M | Premium orchestration over frontier models |
| GPT-6 Astra | $10.00 cache $1.00 | $50.00 | $6.82 +17× | 1.05M | OpenAI's top tier |
| Claude Fable 5.1 | $10.00 cache $0.25 | $50.00 | $6.26 +15× | 1M | Anthropic's top tier |
| Claude Sonnet 5.5 | $2.00 cache $0.10 | $10.00 | $1.29 +3.1× | 1M | Anthropic's mid tier for coding agents |
| DeepSeek V4.1 Flash | $0.30 cache $0.006 | $1.20 | $0.15 −2.7× | 1M | Budget reasoning and coding |
Frequently asked.
What Pareto costs, what you are actually buying, and how the card has moved with each release.
Q · 01 How much does Pareto cost? +
$0.80/M input, $0.03/M cached input and $3.20/M output on the pay-as-you-go API, billed against prepaid credits - or $0.411/M on our 92/8 agentic blend with 82% cache hits. Unbiased also sells personal subscriptions with weekly token allowances; that is a different product, and this page covers only the per-token rate.
Q · 02 What is a blended model? +
Unbiased's description: several LLMs, frontier and open, work on every request in parallel, and their results are synthesized into the one answer you get back. It says Pareto is not a router - it does not pick one model per prompt and never switches models mid-conversation, so your prompt cache survives. It does not name the models in the blend and says the composition can change between releases, so what you are buying is the rate card and the published results, not a fixed set of weights.
Q · 03 Did the price change? +
Twice in under three weeks, and the public changelog records neither change. The Internet Archive's capture of the pricing page on September 17, 2026 shows $1.25 input, $0.15 cached and $6.25 output per 1M under release 26.8. By early September 18 the card read $2.50 / $0.25 / $7.50 under 26.9 - input doubled, output up 20% - and that is what we read on September 24. On October 1 the changelog made 26.10 Preview the current release under the same model string, and by October 4 the card read $0.80 / $0.03 / $3.20: input down 68%, output down 57%, cached input down 88%. The last archived capture before the cut is from September 21, so the exact day of the change is not on record.
Q · 04 How does it score against single frontier models? +
On the slate Unbiased publishes for 26.10 Preview, which it labels preliminary: 69.9 on DeepSWE v1.1, 50.8 on Terminal-Bench 4.0, 49.9 on Humanity's Last Exam (text-only) and 92.4 on GPQA-Diamond. On all four that is below GPT-6 Astra, GPT-6.1 Sol and Claude Sonnet 5.5 in the same table, and above GPT-6 Luna. The claim Unbiased makes is about cost, not rank: it now publishes a mean cost per task - $0.24 on DeepSWE, $0.48 on Terminal-Bench - and says Pareto matches or sets the best score available at its cost. The other models' numbers are taken from their labs and Artificial Analysis rather than rerun, so read the table as a vendor's comparison; the evaluation harness is public if you want to rerun Pareto yourself.
Q · 05 How do I get access? +
Pareto is in private beta: create an account on Unbiased's platform and request access, and a person reviews every signup before the API key works. The API speaks both OpenAI Chat Completions and Anthropic Messages at
https://api.unbiased.ai/v1 with the model string pareto. Pay-as-you-go allows 300 requests per minute with no weekly cap. One quirk from the docs: unknown model ids are not rejected yet, so a typo is served and billed as Pareto.Q · 06 What is the context window, and where does my data go? +
Unbiased publishes no context window, so this page does not state one. Prompts and responses pass through the infrastructure of Circuit & Chisel, the company that builds and runs Unbiased. Its data policy, effective September 24, 2026, says pay-as-you-go prompts and responses are kept no longer than it takes to process the request, or up to 30 days where abuse prevention or security monitoring needs them; subscription plans keep content for up to 30 days. Read it before sending anything sensitive.