Claude Opus 4.8 Fast felt wildly overpriced in one coding session, user says
immersive-matthew · reddit · 2026-07-21
The complaint
The author says Claude Opus 4.8 (Fast) on OpenRouter felt far too expensive for the value delivered. In their coding workflow, the model was only marginally faster than a local Qwen 3.6 27B MTP running on a 4090.
What happened
- They burned through credits after about 22,000 tokens.
- The session cost just under $9 and still did not finish.
- OpenRouter said the prompt would need at least 32,000 tokens, which would have pushed the cost to roughly $13.
- Switching back to local Qwen and continuing the task finished it in about 10 seconds and it tested correctly.
Their takeaway
They argue that Claude’s pricing feels out of touch compared with local inference and other models they tried, including:
- Kimi K3
- GPT 5.6 SOL
- Deep Seek V4 pro
Those alternatives reportedly cost sub-$1 for similar or larger sessions, while giving comparable results for their use case. The author says they won’t keep paying for Claude at that rate, though others may find it worth the cost for different tasks.
More from coding & agent
- Agent verified 'success' after double-voting: a lesson in harness vs model bugs — TejasKumar_ · 2026-10-26
- diginomica: agentic automation hype is overdone, and 'The Hackening' of enterprise security has begun — jonerp · 2026-10-26
- 'Vibe verification': ~60% of AI-written formal specs prove the wrong property — basedjensen · 2026-10-26
- Devin launches Security suite powered by Claude Mythos with $100M+ credits — DevinAI · 2026-10-26
- HackBench injects reward hacks into agent trajectories to stress-test monitors — DanielKhashabi · 2026-10-26
- Hugging Face ships hf CLI updates: deduped cache saves 19GB, faster buckets, retryable GPU jobs — yb2698 · 2026-10-26