Last refreshed 2026-07-26. Next refresh: weekly.
Why use Claude Mythos 5 on Anthropic?
Anthropic offers Claude Mythos 5 with pay-as-you-go pricing at $10.00/1M input tokens. Anthropic is a safety-focused AI research company and creator of the Claude model family, offering API access through the Anthropic Console for text generation, analysis, and coding tasks.
Compare Claude Mythos 5 across 3 providers to find the best fit for your use caseSetup recipe
Python + curlpip install anthropicexport ANTHROPIC_API_KEY=...import anthropic
client = anthropic.Anthropic() # reads ANTHROPIC_API_KEY from env
message = client.messages.create(
model="claude-mythos-5",claude-mythos-5Request example
import anthropic
client = anthropic.Anthropic() # reads ANTHROPIC_API_KEY from env
message = client.messages.create(
model="claude-mythos-5",
max_tokens=1024,
messages=[{"role": "user", "content": "Hello"}]
)
print(message.content[0].text)Gotchas
- Use the versioned model ID, e.g. "claude-sonnet-4-6" or "claude-opus-4-7-20251101".
- The examples expect ANTHROPIC_API_KEY; rename it only if your application config maps the new variable.
Compare Claude Mythos 5 Across Providers
| Provider | Input (per 1M) | Output (per 1M) |
|---|---|---|
| Anthropic | $10.00 | $50.00 |
| AWS Bedrock | — | — |
| GCP Vertex AI | — | — |
Pricing
| Type | Price (per 1M) |
|---|---|
| Input tokens | $10.00 |
| Output tokens | $50.00 |
Capabilities
About Claude Mythos 5
Anthropic's access-gated frontier model for approved Project Glasswing cybersecurity defenders and biomedical research organizations. Shares the same underlying architecture as Claude Fable 5 but operates with safety classifiers lifted in specific domains: cybersecurity safeguards are removed for all Glasswing participants, while biology safeguards are additionally removed for approved biology-track participants. Succeeds Claude Mythos Preview with significantly reduced pricing ($10/$50 per MTok input/output vs. $25/$125 for Mythos Preview), a 1M-token context window, 128k max output tokens, adaptive thinking always on (raw chain of thought never returned), vision, tool use, structured outputs, and the effort parameter for controlling thinking depth.