Claude Mythos Preview vs GPT-5.5 Spud (April 2026)

April 2026 is shaping up to be the biggest month in AI since the original ChatGPT launch. Two frontier models are converging: Anthropic’s Claude Mythos Preview (released April 7, restricted) and OpenAI’s GPT-5.5 “Spud” (pretraining complete, release imminent).

Here’s everything we know about how they compare.

Last verified: April 12, 2026

Status Check

	Claude Mythos Preview	GPT-5.5 Spud
Status	Released (restricted)	Pretraining complete, not yet released
Release date	April 7, 2026	Expected April-June 2026
Access	Project Glasswing (invite-only)	TBD
Pricing	$25/$125 per 1M tokens	TBD
Provider	Anthropic	OpenAI

What We Know About Each

Claude Mythos Preview

Anthropic’s most powerful model, released under restricted access:

• Cybersecurity breakthrough — Finds zero-day vulnerabilities that human researchers missed
• Code exploits — Developed working exploits 181 times in internal benchmarks
• Register control — Achieved near-complete exploit on 29 additional targets
• “Step change” in reasoning according to Fortune’s reporting on the leaked data
• Restricted because Anthropic considers it too powerful for unrestricted release

GPT-5.5 Spud

OpenAI’s next frontier model, internally codenamed “Spud”:

• Pretraining completed around March 24, 2026
• Long-horizon tasks — Designed for complex, multi-step work
• Polymarket odds — 78% chance of April release, 95%+ by June
• Expected to be widely available — OpenAI typically launches broadly
• Competitive pressure — OpenAI wants to ship before Mythos access expands

Capability Comparison (What We Can Infer)

Capability	Mythos Preview	GPT-5.5 Spud (Expected)
Coding	Exceptional (exploit-level)	Strong (GPT-5 family is solid)
Reasoning	”Step change” improvement	Expected major leap
Cybersecurity	Industry-leading	Unknown
General knowledge	Frontier-level	Frontier-level
Context window	Unknown (likely large)	Unknown
Multimodal	Unknown	Likely (GPT-5.4 is multimodal)
Availability	Restricted	Expected broad

The Real Competition

Here’s what most people miss: neither model is the one most developers will use this month.

The practical landscape right now:

Model	Available?	Price	Best for
Claude Opus 4.6	✅ Yes	$15/$75 per 1M	Coding, reasoning, long tasks
GPT-5.4	✅ Yes	$10/$30 per 1M	General tasks, multimodal
Gemini 3.1 Pro	✅ Yes	Varies	Long context, Google ecosystem
Llama 5 600B	✅ Self-host	Free (compute costs)	Open-source, privacy
Claude Mythos Preview	⚠️ Restricted	$25/$125 per 1M	Enterprise security
GPT-5.5 Spud	❌ Not yet	TBD	TBD

Strategic Implications

For Anthropic

Mythos Preview’s restricted release is a bold move. It positions Anthropic as the “responsible” lab that gates dangerous capabilities — but also creates scarcity and FOMO among enterprise customers.

For OpenAI

The pressure to ship Spud before Mythos goes wider is real. OpenAI has historically favored broad launches, which could be their advantage: if Spud is even 80% as capable but 100% available, most developers will choose it.

For Developers

Don’t wait. Build with what’s available now (Opus 4.6, GPT-5.4, Gemini 3.1 Pro). The model you can use today beats the model you might access next month.

Pricing (Known vs Expected)

Model	Input/1M	Output/1M
Claude Mythos Preview	$25.00	$125.00
GPT-5.5 Spud	TBD (likely $15-25)	TBD (likely $60-100)
Claude Opus 4.6	$15.00	$75.00
GPT-5.4	$10.00	$30.00

The Takeaway

Claude Mythos Preview is the more impressive model on paper — Anthropic’s safety concerns alone signal a capability leap. But it’s restricted and expensive.

GPT-5.5 Spud will likely be the more impactful model because OpenAI will make it widely available. The model that changes the industry is the one people can actually use.

For now, build with Opus 4.6 or GPT-5.4. The frontier model wars of April 2026 will shake out over the next few weeks.

Last verified: April 12, 2026