aiaimodel.cards

Compare

Pick up to four language models and read them side by side.

Models · 3 of 4 pickedShow
Anthropic
OpenAI
Google
xAI
Meta
Alibaba
Mistral AI
DeepSeek
ByteDance
Claude Fable 5.1AnthropicGPT-6 AstraOpenAIGemini 3.1 ProGoogle
Specs
Input / 1M$10.00$10.00$2.00
Output / 1M$50.00$50.00$12.00
Cache read / 1M$0.25$1.00$0.20
Context1M1.05M1M
Max output128K128K64K
Knows up toJune 2026April 2026January 2025
ReasoningAdaptive (always on)Reasoning, always on (effort low → max)Dynamic, thinking_level low / medium / high
Default efforthigh—high
Inputstext, image, pdftext, imagetext, image, audio, video, pdf
WeightsClosedClosedClosed
Released1 Sep 20263 Sep 202619 Feb 2026
API model IDclaude-fable-5-1gpt-6-astragemini-3.1-pro-preview
From each card
Built for
  • Coding, knowledge work and problem-solving, including novel mathematical and scientific reasoning
  • Largest gains over Fable 5 in terminal-based scientific and engineering work, computer use, and long-horizon agentic and professional work
  • Matches or exceeds Fable 5 at roughly half the cost per task on agentic coding benchmarks
  • Complex reasoning and coding
  • Computer use and browsing
  • Research and document creation
  • Agentic performance
  • Advanced coding
  • Long context and multimodal understanding
Watch out for
  • Prompt injection gets through on fallback turns
  • Overstates what the user authorized
  • Guesses credentials in computer use
  • Oversteps permissions in agentic coding
  • Out-of-scope attacks in hard cyber tasks
  • Harder to monitor through its reasoning
  • Cyber capability at alert threshold
  • Hallucinations
  • Old knowledge cutoff
SourceSystem Card: Claude Fable 5.1 & Claude Mythos 5.1 ↗GPT-6 Astra System Card ↗Gemini 3.1 Pro Model Card ↗
Benchmarks reported on more than one card (self-reported, settings differ)
Terminal-Bench 4.055.857.9best—
Terminal-Bench-Science 0.152.664.6best—
Humanity's Last Exam (no tools)60.9best—44.4
Humanity's Last Exam (with tools)65.0best57.2—
HealthBench Professionalscore62.164.7best—

Benchmark numbers come from each maker's own card. Makers run them with different settings and harnesses, so treat cross-maker differences as rough.