Openrouter/sakana/fugu Ultra

Sakana Fugu Ultra

Openrouter · LLM · #3 Rank Active
⚖ Compare
94.4
Score
Share

Benchmark Scores

Overall Overall score composition: Reasoning 39% — 100.0; Code 20% — 100.0; Image 12% — 71.4; Text 24% — 100.0
94.4
#6
BenchLM → Airating calculation
Code Airating rank — aggregate for Code (BenchLM / Airating calculation)
100.0
#4
BenchLM
BenchLM · Code — 100.0 (confidence 81%)
Text Airating rank — aggregate for Text (BenchLM / Airating calculation)
100.0
#9
BenchLM
BenchLM · Text — 100.0 (confidence 81%)
Image Airating rank — aggregate for Image (BenchLM / Airating calculation)
71.4
#135
BenchLM
BenchLM · Image — 71.4 (confidence 81%)
Reasoning Airating rank — aggregate for Reasoning (BenchLM / Airating calculation)
100.0
#4
BenchLM
BenchLM · Reasoning — 100.0 (confidence 81%)
Speed Output throughput (tokens/sec) from modelgrep live probes, or catalog estimate as fallback.
0.4
#651
modelgrep · 8 hours ago
Price Normalized API cost (input + output per 1M tokens) from OpenRouter and models.dev.
72.8
#602
models.dev · 8 hours ago

What the rating is based on

Overall score is a weighted composite. Primary data source: BenchLM.

94.4 BenchLM
  • Code 20% 100.0 BenchLM
    • BenchLM · Code 100.0 confidence 81%
  • Text 24% 100.0 BenchLM
    • BenchLM · Text 100.0 confidence 81%
  • Image 12% 71.4 BenchLM
    • BenchLM · Image 71.4 confidence 81%
  • Math 5% No data
  • Reasoning 39% 100.0 BenchLM
    • BenchLM · Reasoning 100.0 confidence 81%

No data for: Math — excluded from the overall score.

How we calculate scores

Description

Sakana Fugu Ultra is engineered specifically for rigorous reasoning and planning tasks, excelling in complex analyses rather than casual conversation. With its impressive context window of 1,000,000 tokens, this model supports extensive data processing, making it particularly valuable for users engaged in deep research, coding, or strategic decision-making. Its multimodal capabilities allow users to integrate visual information into their reasoning, adding another layer of depth to tasks like technical documentation or data visualization. However, potential users should be aware of some trade-offs. The paid API pricing structure, set at $5 for input and $30 for output per 1 million tokens, may pose a barrier for frequent casual users or those with budget constraints. Additionally, while its focus on reasoning is a strength, it may not be the best choice for those seeking a model for light chat or rapid queries, as latency can be an issue in more demanding tasks.

Strengths

  • Exceptional at complex reasoning and planning tasks.
  • Supports multimodal inputs, enhancing analysis with visual data.
  • Large context window allows for deep insights from extensive datasets.
  • Strong performance on coding and technical tasks.

Limitations

  • Paid API model may deter casual users due to costs.
  • Latency can be a concern when handling extensive queries.
  • Less suited for light conversation or quick, simple tasks.

Similar models

All models →

GPT-5.4 Pro

Openrouter

GPT-5.4 Pro by OpenAI is designed specifically for tackling complex reasoning and planning tasks…

100.0 97.5 93.7
$30.00/1M · 1.1M · MM · Reasoning · Tools
Extensive context window supports large-scale reasoning tasks.
⚖ Compare View details →

GPT-5.3 Codex

Openrouter

GPT-5.3 Codex is a coding-oriented endpoint from OpenAI, specifically tailored for developers who…

100.0 96.9 95.8
$1.75/1M · 400K · MM · Reasoning · Tools
Optimized for complex coding tasks and repository management.
⚖ Compare View details →

Grok-4.1

SpaceXAI

Grok-4.1 from xAI features a staggering context window of 2,000,000 tokens, making it the go-to…

99.4 97.8 97.6
$0.20/1M · 2M · MM · Tools
Handles long documents and complex datasets smoothly.
⚖ Compare View details →

Anthropic Claude Fable 5

Openrouter

Anthropic Claude Fable 5 is specifically tailored for complex reasoning and planning tasks…

100.0 100.0 90.0
$10.00/1M · 1M · MM · Reasoning · Tools
Designed for complex reasoning and strategic planning tasks
⚖ Compare View details →

Qwen 3.7 Max

Openrouter

Qwen 3.7 Max is specifically engineered for complex reasoning and planning tasks, making it an…

94.8 93.4 91.6
$1.48/1M · 1M · Reasoning · Tools
Extensive context window of 1,000,000 tokens for deep analysis.
⚖ Compare View details →

Z.ai: GLM-5.2

Openrouter

Z.ai: GLM-5.2 distinguishes itself by offering open weights, allowing users to self-host their AI…

98.0 91.2 89.3
$0.09/1M · 1M · Reasoning · Tools
Open weights allow for self-hosting and customization.
⚖ Compare View details →

Claude Mythos 5

Anthropic

Claude Mythos 5 from Anthropic represents a significant advancement in AI technology, designed to…

100.0 100.0 100.0
$10.00/1M · 1M · MM · Reasoning · Tools
Extensive context window for handling large datasets
⚖ Compare View details →

Claude Sonnet 5

Openrouter

Claude Sonnet 5 by Anthropic is a powerful AI model designed for a variety of text-based tasks…

100.0 97.9 86.1
$2.00/1M · 1M · MM · Reasoning · Tools
Impressive reasoning capabilities for complex tasks.
⚖ Compare View details →