Unbiased releases Pareto 26.9: 262K-context composite model at $2.50/$7.50 per 1M tokens

Unbiased has released Pareto 26.9 (unbiased/pareto-26.9), a multimodal composite model with a 262K-token context window, priced at $2.50 per 1M input tokens and $7.50 per 1M output tokens. The OpenRouter listing gives a release date of Oct 13, 2026.

Specifications

Spec	Pareto 26.9
Developer	Unbiased
Type	Multimodal composite model
Context window	262K tokens
Input price	$2.50 / 1M tokens
Output price	$7.50 / 1M tokens
Cache read price	$0.25 / 1M tokens
Released	Oct 13, 2026
Parameter count	Not disclosed
Training cutoff	Not disclosed

According to the listing, Pareto is "built for research, coding, and agentic workflows," and delivers "frontier-level performance across a broad range of general-purpose tasks." That is a company claim. The listing publishes no benchmark scores, so the performance claim cannot be verified from the available information.

Serving performance

OpenRouter lists Unbiased as the only provider for this model. Reported P50 figures:

• Latency: 2.42 seconds
• Throughput: 76 tokens per second
• OpenRouter availability: 94.73%

OpenRouter notes it can reroute requests to another healthy provider when an upstream error occurs, if request filters allow it. With a single listed provider for this model, that fallback has no alternative endpoint to use. The listing also says there is not yet enough data on apps using the model.

The Pareto family

Unbiased lists several related entries:

• Pareto 26.9: the stable, pinned version covered here.
• Pareto: a separate listing with the same 262K context and $2.50 / $7.50 pricing.
• Pareto Latest: an alias that always redirects to the newest model in the family. The listing shows 262K context and no price.
• Pareto 26.10 Preview: a preview of the next version with 1.0M-token context at $0.80 input / $3.20 output per 1M tokens. Unbiased says it "may change without notice" and recommends pareto-26.9 for stable behavior.

The listing does not describe what the "composite" architecture consists of, such as underlying models, routing logic, or modality coverage for inputs and outputs. It also does not say whether Pareto 26.9 has its own weights.

What this means

The notable detail is the preview, not the release. Pareto 26.10 Preview offers roughly 3.8x the context window of 26.9 at about 68% lower input pricing and 57% lower output pricing. If those terms hold at general availability, 26.9 becomes the premium-priced legacy option within weeks of launch. That makes it hard to justify for anything except workloads that need pinned behavior.

The pinned-version and preview structure suggests Unbiased is iterating quickly and expects developers to pin versions. Teams building agentic or coding pipelines should avoid the Pareto Latest alias in production, since it will silently move behavior to new versions.

Until Unbiased publishes benchmarks, architecture details, and training information, Pareto 26.9 can only be evaluated through direct testing. Single-provider availability at 94.73% is also a consideration for production workloads.