Model
Fugu Ultra
Quality-first multi-agent model for hard research, analysis, and competitions
Compare providers (sorted by listed input price)
11 providers list Fugu Ultra. The cheapest full-context option is NanoGPT at $5/$30 per M.
Best price history
Daily lowest published price across live providers. A lower line is better for the buyer.
- Best input /M
- $5▲ 11.1%
- Best output /M
- $30▲ 11.1%
- Provider coverage
- 12
- 12 at the start · 24 snapshots
Price badge
https://www.vaanalytics.in/badge/sakana/fugu-ultra.svgLive SVG, regenerated on every hourly sync — paste it into a README to always show the current cheapest listed price. Updates when prices change; no build step needed on your side.
Recent changes
Changelog →- repricedvia nano-gpt · Sep 13, 2026
- input: $5.25 → $5
- output: $31.50 → $30
- cache_read: $0.525 → $0.5
- contextvia merge-gateway · Sep 13, 2026
- output: 250K → 128K tokens
- capabilityvia requesty · Sep 2, 2026
- reasoning: false → true
- repricedvia requesty · Aug 28, 2026
- input: $4.50 → $5
- output: $27 → $30
- cache_read: $0.45 → $0.5
- tiers: [object Object] → [object Object]
Benchmarks
- SWE Bench Pro ↗report ↗Unclassified source73.7
- Terminal Bench 2.1 ↗report ↗Unclassified source82.1
- LiveCodeBench ↗report ↗Unclassified source93.2
- LiveCodeBench Pro ↗report ↗Unclassified source90.8
- Humanity’s Last Exam ↗report ↗Unclassified source50
- CharXiv Reasoningreport ↗Unclassified source86.6
- GPQA Diamond ↗report ↗Unclassified source95.5
- SciCodereport ↗Unclassified source58.7
- τ3 Bankingreport ↗Unclassified source20.6
- Long Context Reasoningreport ↗Unclassified source73.3
- MRCRv2report ↗Unclassified source93.6
- CTI-REALMreport ↗Unclassified source69.4
Scores a lab published about its own model are claims, not measurements. Rankings elsewhere on this site use independently measured scores only.
Prices are per million tokens (USD). “—” means the provider does not publicly list a price for this model. Data from models.dev; verify with the provider before purchasing.