The AI Gateway built for Privacy & Security 🐳 0% Markup. BYOK. Adaptive Routing. Firewall. Guardrails. Observability. Cyber. → discord.gg/YEubt8enRA
- Tweets 4,419
- Following 23
- Followers 22,807
- Likes 1,444
Pinned Tweet
🚨 Emergency release: OrcaCyber Nano 1.0 is now open-weight.
Part of our OrcaCyber model series: Nano, Zero, Ultra.
Based on our early evals, Nano is the most capable local model in its under-16GB class. 27B parameters. 262K context. Vision. Fully offline.
Early benchmarks:
AIME 2025: 96.23%
MATH-500: 98.60%
GSM8K: 97.50%
MBPP+: 84.66%
IFEval: 82.81%
Built for AI defense, vulnerability research, code security, and incident response.
We're shipping early because defenders can't wait. Cyber-specific evals coming next.
huggingface.co/orcarouter/Or…
58
140
1,646
526,380
The most underrated model optimization is generating fewer tokens to solve harder problems.
Early observations on DeepSeek V4.1 Flash 1017 vs. 0910 based on our internal infra data:
~30% lower TTFT.
Shorter reasoning traces and outputs, eliminating redundant tokens.
Lower cost per completed task, despite identical per-token pricing.
Better intelligence: Terminal-Bench improves from 31.2 → 42.3; DeepSWE from 74.2 → 77.0.
Our research team investigated these changes. Our leading hypothesis is that DeepSeek very likely applied self-distillation to compress redundant reasoning while preserving or improving capability. The exact training method remains unconfirmed. Deepseek team seems to optimize for more intelligence per generated token.
3
3
91
6,399
🚨 Emergency release: OrcaCyber Nano 1.0 is now open-weight.
Part of our OrcaCyber model series: Nano, Zero, Ultra.
Based on our early evals, Nano is the most capable local model in its under-16GB class. 27B parameters. 262K context. Vision. Fully offline.
Early benchmarks:
AIME 2025: 96.23%
MATH-500: 98.60%
GSM8K: 97.50%
MBPP+: 84.66%
IFEval: 82.81%
Built for AI defense, vulnerability research, code security, and incident response.
We're shipping early because defenders can't wait. Cyber-specific evals coming next.
huggingface.co/orcarouter/Or…
58
140
1,646
526,380
OrcaCyber Nano 1.0 just ranked #1 among 21 models in an independent community evaluation for 16GB GPUs, scoring 89.8/100.
72 tasks across 7 categories: Coding, Reasoning, Instruction Following, Web Search, Agent, Vision, and Long Context.
27B parameters. 13GB model. 60 tokens/sec on a single V100 16GB.
Full benchmarks: github.com/Abdelhayali/V100-…
840
OrcaCyber Nano 1.0 just ranked #1 among 21 models in an independent community evaluation for 16GB GPUs, scoring 89.8/100.
72 tasks across 7 categories: Coding, Reasoning, Instruction Following, Web Search, Agent, Vision, and Long Context.
27B parameters. 13GB model. 60 tokens/sec on a single V100 16GB.
Full benchmarks: github.com/Abdelhayali/V100-…
Full eval results are in.
OrcaCyber Nano 1.0 scores 78.4% on SWE-bench Verified, nearly matching DeepSeek V4 Flash's reported 79.0%.
A 27B dense open-weight model designed to run locally under 16GB for cyber defense.
huggingface.co/orcarouter/Or…
7
4
171
8,799
Full eval results are in.
OrcaCyber Nano 1.0 scores 78.4% on SWE-bench Verified, nearly matching DeepSeek V4 Flash's reported 79.0%.
A 27B dense open-weight model designed to run locally under 16GB for cyber defense.
huggingface.co/orcarouter/Or…
2
1,649
Full eval results are in.
OrcaCyber Nano 1.0 scores 78.4% on SWE-bench Verified, nearly matching DeepSeek V4 Flash's reported 79.0%.
A 27B dense open-weight model designed to run locally under 16GB for cyber defense.
huggingface.co/orcarouter/Or…
🚨 Emergency release: OrcaCyber Nano 1.0 is now open-weight.
Part of our OrcaCyber model series: Nano, Zero, Ultra.
Based on our early evals, Nano is the most capable local model in its under-16GB class. 27B parameters. 262K context. Vision. Fully offline.
Early benchmarks:
AIME 2025: 96.23%
MATH-500: 98.60%
GSM8K: 97.50%
MBPP+: 84.66%
IFEval: 82.81%
Built for AI defense, vulnerability research, code security, and incident response.
We're shipping early because defenders can't wait. Cyber-specific evals coming next.
huggingface.co/orcarouter/Or…
20
24
306
24,189
OrcaCyber Nano 1.0 delivers performance on par with GPT-5, GLM-5.3 Flash, and DeepSeek V4 Flash on select benchmarks.
At a fraction of the size. Runs locally on a single 24GB GPU. Open weights. Built for defensive cyber.
1
1
34
3,988
Following the OpenCode live hack incident, one thing is clear: everyone could be a target.
We're now offering FREE priority access to OrcaCyber models for anyone actively under cyberattack.
Defend. Investigate. Remediate.
Apply now: orcarouter.ai/models?tier=cy…
6
8
121
95,553
🚨 Emergency release: OrcaCyber Nano 1.0 is now open-weight.
Part of our OrcaCyber model series: Nano, Zero, Ultra.
Based on our early evals, Nano is the most capable local model in its under-16GB class. 27B parameters. 262K context. Vision. Fully offline.
Early benchmarks:
AIME 2025: 96.23%
MATH-500: 98.60%
GSM8K: 97.50%
MBPP+: 84.66%
IFEval: 82.81%
Built for AI defense, vulnerability research, code security, and incident response.
We're shipping early because defenders can't wait. Cyber-specific evals coming next.
huggingface.co/orcarouter/Or…
5
597
API Access to OrcaCyber Zero orcarouter.ai/models/orca/or…
14
4,667
OrcaCyber Zero is one of the strongest cyber models out there
We're excited to introduce OrcaCyber Zero 1.5 — our latest frontier cybersecurity model, built to discover critical vulnerabilities, investigate complex attack paths, and validate real-world exploitability.
Updated weights. New benchmark results:
• 95.8% — CVE-Bench (23/24 evaluable tasks)
• 100% — Cybench (39/39, unrestricted)
• 93.9% — HumanEval+
Its predecessor, OrcaCyber Zero 1.0, has already helped protect some of the world's largest software companies, governments, blockchain networks, and cybersecurity firms from real-world threats.
It also powered our autonomous OrcaCyber harness, which achieved 98%+ pass@1 on CyberGym, ranking among the top 3 globally.
What's different about OrcaCyber Zero 1.5?
🔹 Find what others miss. Designed to uncover previously unknown vulnerabilities, including RCE, sandbox escapes, authentication bypasses, privilege escalation, and complex attack chains.
🔹 Go beyond detection. Reasons through attack paths, challenges its own hypotheses, and prioritizes vulnerabilities with demonstrable exploitability and real-world impact.
🔹 Built for autonomous security agents. 1M-token context window, native tool calling, and extended reasoning to investigate large codebases and complex attack surfaces.
The goal isn't more vulnerability reports. It's more real vulnerabilities discovered, validated, and fixed before attackers exploit them.
Available now exclusively on OrcaRouter:
orcarouter.ai/models/orca/or…
2
10
951
Every company will get hacked. It’s a matter of when, not if.
Apply for OrcaCyber model access to get prepared: orcarouter.ai/models?tier=cy…
5
4
83
5,793
Open models and community engineering matter.
A developer got our Qwen3.8 Flash Next Uncensored running on 3 aging AMD MI50 GPUs.
Initially: 0.5 tokens/sec.
After optimization: ~10 tokens/sec.
20× faster. Same GPUs. Same model weights.
The biggest surprise? Reducing CPU threads from 32 to 8–12 dramatically improved performance.
Sometimes the bottleneck isn't the GPU. It's how you use it. This is the power of open-source AI.
huggingface.co/orcarouter/Qw…
7
2
97
6,692
The OpenCode team are great builders, and we have huge respect for them. Security incidents can happen to anyone, especially at scale.
We've faced similar attacks at OrcaRouter.
This is a reminder that continuous security code reviews and pentesting are essential, but so is protecting BYOK credentials.
At OrcaRouter, BYOK keys are encrypted at rest with AES-256-GCM.
The entire AI ecosystem needs stronger security. We're rooting for the OpenCode team to get through this. 💪
We had a security breach last night related to a large scale distillation attack that impacted 1.4% of OpenCode Console users
We have notified them all this morning via email. If you did not get an email you were not impacted, it was a subset of users who had external providers linked
The incident is ongoing so we cannot share more details at this point but we are working with Anthropic, OpenAI, and Google to remediate
4
9
364
16,355
Stolen AI API keys are becoming the new stolen credit cards. Three ways attackers can monetize them:
Model distillation: Harvest millions of frontier-model responses to train competing models, at someone else's expense.
API resale: Sell cheap access to GPT, Claude, and Gemini through unauthorized proxies.
Compute theft: Run massive inference workloads while victims foot the bill.
One stolen key can generate enormous charges before anyone notices. If you use BYOK anywhere: rotate exposed keys, set spending limits, and monitor usage spikes.
2
14
1,518
DeepSeek V4.1 Flash FREE is also live: orcarouter.ai/models/deepsee…
Terms here: orcarouter.ai/offers
DeepSeek V4.1 Flash has been updated with the latest model weights on OrcaRouter! 🚀
Same model name, same API. No changes required.
orcarouter.ai/models/deepsee…
17
32
722
41,318