O
OpenAI
Released
Version6 Astra
- Parameters
- Not disclosed
- Context window
- 1.05M
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
Best at computer use and long-horizon agents; Sol and Luna siblings available
O
OpenAI
Released
Version2.5(Flare / Sunburst)
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- —
- API
- Web
- Functions
- JSON
- OpenAI API
- Caching
Strong text rendering and multi-turn editing, 50% faster generation
O
OpenAI
Released
Version2.1
- Parameters
- Not disclosed
- Context window
- 128K
- Modality
- Text · Audio
- License
- —
- Knowledge cutoff
- 2024-09-30
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
The first real-time voice dialogue model with GPT-5-level reasoning
O
OpenAI
Released
Version3 large
- Parameters
- Not disclosed
- Context window
- 8191
- Modality
- Text
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
3072 dimensions; the default choice for English retrieval
A
Anthropic
Released
VersionOpus 5.5
- Parameters
- Not disclosed
- Context window
- Not disclosed
- Modality
- Text · Image
- License
- —
- Knowledge cutoff
- 2026-06-01
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
The most capable Opus, priced 40% below Opus 5
A
Anthropic
Released
VersionFable 5.1
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image
- License
- —
- Knowledge cutoff
- 2026-06-01
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
Best for coding and research; Mythos 5.1 is its restricted sibling
A
Anthropic
Released
VersionSonnet 5
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
The most agentic Sonnet, at a strong price-performance ratio
G
Google DeepMind
Released
Version3.1 Pro
- Parameters
- Not disclosed
- Context window
- Not disclosed
- Modality
- Text · Image · Audio · Video
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
Flagship Pro tier leading 12 of 18 benchmarks
G
Google DeepMind
Released
Version3.8 Flash
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image · Audio · Video
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
The current workhorse, with adjustable effort to balance cost and quality
G
Google DeepMind
Released
Version3.8 Live / Live Extended Thinking
- Parameters
- Not disclosed
- Context window
- Not disclosed
- Modality
- Audio · Text · Image
- License
- —
- API
- Reasoning
- Functions
- JSON
- OpenAI API
First on the speech-quality leaderboard; speaks while it reasons
G
Google DeepMind
Released
VersionPro(Gemini 3 Pro Image)
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
Best-in-class 4K and multilingual text rendering
G
Google DeepMind
Released
Version3.1
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
Native synchronized audio with cinematic camera control
G
Google DeepMind
Released
Version3.5
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
Generates complete 3-minute songs with vocals, at about $0.08 each
G
Google DeepMind
Released
Version4
- Parameters
- Not disclosed
- Context window
- Not disclosed
- Modality
- Text · Image
- License
- Gemma Terms of Use
- API
- Web
- Reasoning
- Functions
- JSON
Flagship open weights, emphasizing intelligence density per parameter
G
Google DeepMind
Released
Version2
- Parameters
- Not disclosed
- Context window
- 8192
- Modality
- Text · Image · Audio · Video
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
Unified multimodal vector space, adjustable from 128 to 3072 dimensions
X
xAI
Released
Version4.7
- Parameters
- Not disclosed
- Context window
- 500K
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
Built for coding and agents, with low hallucination rates and aggressive pricing
X
xAI
Released
Version2.0
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
Inpainting with multiple reference images, about $0.04 per image
X
xAI
Released
Version1.5
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
6-second 720p with native audio, rendered in about 25 seconds
M
Meta
Released
Version1.3
- Parameters
- Not disclosed
- Context window
- 1.05M
- Modality
- Text · Image · Video · Audio
- License
- —
Built for long agentic tasks; output priced below peer flagships
M
Meta
Released
Version30B
- Parameters
- 29.6B
- Context window
- 131K
- Modality
- Text · Image
- License
- Llama Community License
- Knowledge cutoff
- 2026-01-04
Apache 2.0; a local agent that runs on 24GB of VRAM
N
NVIDIA
Released
Version3 Ultra 550B-A55B
- Parameters
- 550B (55B active)
- Context window
- Not disclosed
- Modality
- Text
- License
- NVIDIA Open Model License
- API
- Reasoning
- Functions
- JSON
- OpenAI API
The largest US open-weight reasoning model
N
NVIDIA
Released
Version3.5 Lightning
- Parameters
- 30B (3B active)
- Context window
- 1M
- Modality
- Text
- License
- NVIDIA Open Model License
Runs on a single laptop GPU; fastest agent throughput
N
NVIDIA
Released
Version2.0
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- NVIDIA Open Model License
Generates a walkable 3D world from one image, exportable as mesh or 3DGS
M
Microsoft
Released
VersionThinking-1
- Parameters
- Not disclosed
- Context window
- Not disclosed
- Modality
- Text
- License
- —
- API
- Reasoning
- Functions
- OpenAI API
Enterprise mid-size reasoning, trained in-house rather than distilled
M
Microsoft
Released
Version2.5
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
Text-to-image generation and editing, plus a faster Flash tier
M
Microsoft
Released
Version2
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Audio
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
Low-latency, highly expressive speech synthesis
A
Amazon
Released
Version2 Lite
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image · Video
- License
- —
- API
- Web
- Reasoning
- Functions
- OpenAI API
Bedrock's value workhorse, with a built-in code interpreter
A
Amazon
Released
Version2 Sonic
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Audio
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
Low-latency real-time speech-to-speech
A
Apple
Released
VersionAFM 3(Core / Core Advanced)
- Parameters
- 3B / 20B (1–4B active)
- Context window
- Not disclosed
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
An on-device multimodal model with zero cost and zero upload locally
A
Allen Institute for AI
Released
Version3.1 Think 32B
- Parameters
- 32B
- Context window
- 64K
- Modality
- Text
- License
- Apache-2.0
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
Full training pipeline open-sourced; one of the strongest open thinking models
I
IBM
Released
Version4.1
- Parameters
- Not disclosed
- Context window
- 32K
- Modality
- Text · Image · Audio
- License
- Apache-2.0
- API
- Web
- Reasoning
- Functions
- OpenAI API
Apache 2.0 enterprise-grade, with speech and embedding variants
P
Perplexity
Released
VersionSonar 2
- Parameters
- Not disclosed
- Context window
- Not disclosed
- Modality
- Text
- License
- —
- API
- Web
- Functions
- OpenAI API
Built-in web retrieval and citations, purpose-built for search
L
Liquid AI
Released
VersionLFM2.5-2.6B
- Parameters
- 2.69B
- Context window
- 131K
- Modality
- Text
- License
- LFM Open Weights License
An on-device agent matching models four times its size
C
Cartesia
Released
VersionSonic 3.6
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Audio
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- Caching
First audio under 90ms; tops the real-time speech quality leaderboard
E
ElevenLabs
Released
Versionv3
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Audio
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
Best emotion control; the first choice for long-form audiobooks and dubbing
R
Runway
Released
VersionGen-4.5
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- Caching
Prompt-driven multi-shot camera movement control
L
Luma AI
Released
VersionRay 3.2
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- Caching
Up to 16 keyframes with HDR and EXR output
M
Midjourney
Released
VersionV8.2
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
Best-in-class aesthetics and photorealism; no official API
S
Suno
Released
Versionv5.5
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text
- License
- —
- API
- Web
- Reasoning
- Functions
- JSON
- Caching
One-click full songs; the most natural vocals and arrangement
O
OpenAI
Released
Version6.1
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image
- License
- —
- Knowledge cutoff
- 2026-04-30
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
The mid-tier workhorse of the GPT-6 family announced at DevDay 2026: agentic coding and computer use close to GPT-6 Astra at a fifth of the price
O
OpenAI
Released
Version6
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image
- License
- —
The ultra-low-cost tier of the GPT-6 family at $0.10/$0.50 per million tokens, aimed at large-scale lightweight tasks
G
Google DeepMind
Released
Version4
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image · Video
- License
- —
Google's new flagship and the first Gemini 4 model: long-horizon reasoning plus cyber-defence, with the output ceiling raised to roughly 1M tokens; released in stages under the Fairwind programme
A
Anthropic
Released
Version5.5
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image
- License
- —
- Knowledge cutoff
- 2026-06-01
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
More than 30% faster than Sonnet 5 and up to 30% cheaper per task, with Terminal-Bench 4.0 jumping from 10.3% to 70.6%; already used in the free Claude tier
I
IBM
Released
Version4.2
- Parameters
- 3B / 8B / 30B (dense)
- Context window
- 128K (30B extends to 512K)
- Modality
- Text
- License
- Apache-2.0
- API
- Reasoning
- Functions
- OpenAI API
An open dense reasoning family built for enterprise agents, with a native thinking mode and reasoning-enhanced tool calling; the 8B and 30B sizes were RL-trained in real environments
M
Meta
Released
Version4 Maverick
- Parameters
- 400B MoE (17B active)
- Context window
- 1M
- Modality
- Text · Image
- License
- Llama 4 Community License
- Knowledge cutoff
- 2024-08-01
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
Meta’s first natively multimodal open-weights MoE — 128 experts, 17B active parameters.
M
Meta
Released
Version4 Scout
- Parameters
- 109B MoE (17B active)
- Context window
- 10M
- Modality
- Text · Image
- License
- Llama 4 Community License
- Knowledge cutoff
- 2024-08-01
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
A 10M-token context window; runs on a single H100 once quantised.
M
Microsoft
Released
Version4
- Parameters
- 14B
- Context window
- 16K
- Modality
- Text
- License
- MIT
- Knowledge cutoff
- 2024-06-01
A quality benchmark for small models: trained on synthetic textbook data, strong at maths and reasoning.
O
OpenAI
Released
Versionlarge-v3
- Parameters
- 1.55B
- Context window
- 30 s audio window
- Modality
- Text · Audio
- License
- MIT
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
The open speech-recognition benchmark: transcription and translation across 99 languages.
L
LLaVA Team
Released
Version1.6 / 34B
- Parameters
- 34B
- Context window
- 4K
- Modality
- Text · Image
- License
- Apache-2.0
- Web
- Reasoning
- Functions
- OpenAI API
The defining open visual-instruction model — the classic CLIP + LLM baseline.
N
Nomic AI
Released
Versionv1.5
- Parameters
- 137M
- Context window
- 8192
- Modality
- Text
- License
- Apache-2.0
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
A fully reproducible text embedding model with an 8192-token context.
M
Microsoft
Released
Versionmistral-7b-instruct
- Parameters
- 7B
- Context window
- 32K
- Modality
- Text
- License
- MIT
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
An open embedding model built on Mistral-7B, strong at long-context retrieval.
D
Databricks
Released
VersionInstruct / 132B
- Parameters
- 132B MoE (36B active)
- Context window
- 32K
- Modality
- Text
- License
- Databricks Open Model License
- Web
- Reasoning
- Functions
- OpenAI API
Databricks’ open MoE model, strong at code and SQL generation.
S
Snowflake
Released
VersionInstruct / 480B
- Parameters
- 480B MoE (17B active)
- Context window
- 4K
- Modality
- Text
- License
- Apache-2.0
An enterprise-grade open MoE under Apache-2.0, notable for code and SQL.
O
OpenAI
Released
Version120b
- Parameters
- 117B MoE (5.1B active)
- Context window
- 128K
- Modality
- Text
- License
- Apache-2.0
- Knowledge cutoff
- 2024-06-01
- API
- Web
- Reasoning
- Functions
- JSON
- OpenAI API
OpenAI’s first open-weights model; runs on a single 80GB GPU.
A
Anthropic
Released
Superseded by Haiku 5.5 on 2026-10-07; still callable
VersionHaiku 4.5
- Parameters
- Not disclosed
- Context window
- 200K
- Modality
- Text · Image
- License
- —
- Knowledge cutoff
- 2025-02-28
- API
- Reasoning
- Functions
- JSON
- OpenAI API
- Caching
Anthropic’s fastest and cheapest tier, aimed at real-time and high-concurrency agent workloads.
W
Writer
Released
VersionX5
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image
- License
- —
- API
- Reasoning
- Functions
- JSON
An enterprise long-context model with a 1M-token window, built around multi-agent workflows.
R
Reka AI
Released
VersionCore
- Parameters
- Not disclosed
- Context window
- 128K
- Modality
- Text · Image · Audio · Video
- License
- —
- API
- Functions
- JSON
- OpenAI API
A natively multimodal frontier model: text, image, video and audio understanding behind one API.
G
Google Cloud
Released
VersionChirp 3
- Parameters
- Not disclosed
- Context window
- —
- Modality
- Audio · Text
- License
- —
Google Cloud’s flagship speech-recognition model, with streaming and dynamic batching.
A
AssemblyAI
Released
VersionUniversal-3.5 Pro
- Parameters
- Not disclosed
- Context window
- —
- Modality
- Audio · Text
- License
- —
High-accuracy transcription with speaker diarisation and domain modes such as medical.
D
Deepgram
Released
VersionNova-3
- Parameters
- Not disclosed
- Context window
- —
- Modality
- Audio · Text
- License
- —
Leading low-latency streaming recognition at a 5.26% word error rate, with medical and finance lines.
I
Ideogram
Released
Version3.0
- Parameters
- Not disclosed
- Context window
- —
- Modality
- Text · Image
- License
- —
The benchmark for text rendered inside images; strong at layout and branded content.
R
Recraft
Released
VersionV3
- Parameters
- Not disclosed
- Context window
- —
- Modality
- Text · Image
- License
- —
- API
- Functions
- JSON
- OpenAI API
- Caching
A design-oriented image model with a rare native SVG (vector) output mode.
H
Hedra
Released
VersionCharacter 3
- Parameters
- Not disclosed
- Context window
- —
- Modality
- Text · Image · Audio · Video
- License
- —
One portrait plus audio yields high-fidelity talking and singing video, up to 10 minutes long.
P
Pika
Released
Version2.5
- Parameters
- Not disclosed
- Context window
- —
- Modality
- Text · Image · Video
- License
- —
Effects-first short-video generation, with Pikaffects object transformations and more.
M
Microsoft
Released
VersionNeural HD 2.5
- Parameters
- Not disclosed
- Context window
- —
- Modality
- Text · Audio
- License
- —
An enterprise speech platform: synthesis, recognition, translation and custom voices.
A
Amazon
Released
VersionStandard / Neural / Generative / Long-Form
- Parameters
- Not disclosed
- Context window
- —
- Modality
- Text · Audio
- License
- —
AWS-managed TTS with four quality tiers: standard, neural, generative and long-form.
T
TypeSafe AI
Released
Versionjev-1.13.0
- Parameters
- —
- Context window
- 64k tokens per request
- Modality
- Text
- License
- —
- API
- Web
- Reasoning
- Functions
TypeSafe AI's System One decision model: no text output, just typed answers with calibrated probabilities; the SDK and Agent Skill are MIT-licensed.
C
Cloudflare
Released
Version27B
- Parameters
- 27B (frozen Qwen3.8-27B + routing head + LoRA)
- Context window
- 65536
- Modality
- Text · Image · Video
- License
- Apache-2.0
- API
- Web
- Reasoning
- Functions
Cloudflare’s own decision model, API-compatible with Jev and able to read images too.
C
Cloudflare
Released
Version9B
- Parameters
- 9B (frozen Qwen3.5-9B + routing head + LoRA)
- Context window
- 65536
- Modality
- Text · Image · Video
- License
- Apache-2.0
- API
- Web
- Reasoning
- Functions
The 9B lightweight Clef: a decision in about 40ms, the cheapest open option.
C
Convai Innovations
Released
Version421M
- Parameters
- ~421M
- Context window
- 512
- Modality
- Text
- License
- Apache-2.0
- API
- Web
- Reasoning
- Functions
An open decision model at roughly 400M parameters that self-hosts on CPU and at the edge.
P
Perplexity
Released
Versionv1-27b
- Parameters
- 27B
- Context window
- —
- Modality
- Text · Image
- License
- Apache-2.0
- API
- Web
- Reasoning
- Functions
Perplexity’s multimodal decision model: probabilities from text or images, with open weights.
I
Inception
Released
VersionMercury Decide
- Parameters
- —
- Context window
- 33k
- Modality
- Text
- License
- —
- API
- Web
- Reasoning
- Functions
Inception’s decision endpoint using Jev’s question types, free to try.
A
Anthropic
Released
Announced 2026-10-07; GA in the Claude API
Version5.5
- Parameters
- —
- Context window
- 1M
- Modality
- Text · Image
- License
- —
- Knowledge cutoff
- 2026-06-01
One of the four current Claude models; retirement commitment no earlier than 2027-10-07
A
Anthropic
Preview
Not publicly available; requires the LSVP verification program
VersionMythos(5.1 代)
- Parameters
- —
- Context window
- 1M
- Modality
- Text · Image
- License
- —
Listed as accessible under the Life Sciences Verification Program; no standalone launch
A
Anthropic
Unconfirmed
Unannounced; leaked via cloud service config
Version5.5
- Parameters
- —
- Context window
- —
- Modality
- Text · Image
- License
- —
Unannounced; found in regional cloud service records. Covered in our news #28
O
OpenAI
Preview
Gated to authorized security researchers
Version5.6 Cyber
- Parameters
- —
- Context window
- —
- Modality
- Text
- License
- —
Listed under OpenAI Daybreak; requires authorization
O
OpenAI
Preview
Requires institutional approval
VersionRosalind
- Parameters
- —
- Context window
- —
- Modality
- Text
- License
- —
OpenAI docs list it as available to approved institutions
O
OpenAI
Unconfirmed
Unannounced; surfaced via blind arena testing
VersionRed / Green / Alpha
- Parameters
- —
- Context window
- —
- Modality
- Image
- License
- —
Unannounced; spotted in blind arena tests. Covered in our news #32
O
OpenAI
Released
Generally available Oct 26, 2026 in ChatGPT Work, Codex and API (gpt-6.1-terra)
Version6.1
- Parameters
- —
- Context window
- 1.05M
- Modality
- Text · Image
- License
- —
- Knowledge cutoff
- 2026-06-01
- Web
- Reasoning
- Functions
- Caching
Supports computer use, web search and functions; reasoning effort none–max; knowledge cutoff Jun 2026
G
Google DeepMind
Preview
Research preview with a limited group of trusted testers; AI Ultra subscribers and early access partners are next
Version4
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image · Video · Audio
- License
- —
Built on Gemini Omni plus Gemini reasoning models; audio and dialogue generated on the fly
G
Google DeepMind
Released
Generally available since Nov 10, 2026; the Oct 28 launch was limited to cyber defenders and Fairwind partners
Version4 Carbon
- Parameters
- Not disclosed
- Context window
- Not disclosed
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- Caching
Ranks #1 on Artificial Analysis Intelligence Index v5.0 (59); SEC-bench Pro 88.2%, CWE-bench 71.0%
O
OpenAI
Released
Live since Nov 4, 2026 across ChatGPT, Codex and the API, on every subscription tier
Version2.6(Flare / Sunburst)
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image
- License
- —
Flare trades for speed, Sunburst for detail and cross-edit consistency; adds complexity and quality controls
O
OpenAI
Preview
No self-serve access; available only through the Daybreak trusted-access programme (Daybreak Red)
Version6 Cyber
- Parameters
- Not disclosed
- Context window
- Not disclosed
- Modality
- Text
- License
- —
ExploitGym 48.5% vs 32.1% for GPT-5.6 Cyber; classified Critical under the Preparedness Framework
G
Google DeepMind
Released
Generally available Nov 5, 2026 on the Gemini API, Google Cloud, Google Vids and Gemini for business
Version1.2 Flash
- Parameters
- Not disclosed
- Context window
- N/A
- Modality
- Text · Image · Video · Audio
- License
- —
First generally available Gemini Omni release; supports character, product and style references with a SynthID watermark
M
Meta
Released
Announced by Meta AI on Nov 9, 2026
VersionEmber
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image
- License
- —
1M context, aimed at high-frequency repetitive agent subtasks while larger models do the judgement
A
Anthropic
Released
Available to all users since Nov 9, 2026, with cyber and biology safeguards
VersionFable 5.2
- Parameters
- Not disclosed
- Context window
- Not disclosed
- Modality
- Text · Image
- License
- —
- API
- Web
- Reasoning
- Functions
- Caching
FrontierCode 2.0 Xhigh 49.1%, highest of any released model; HLE with tools 74.2%; can serve as Advisor or Orchestrator
A
Anthropic
Preview
Requires approval through the Cyber or Life Sciences Verification Program; no self-serve access
VersionMythos 5.2
- Parameters
- Not disclosed
- Context window
- Not disclosed
- Modality
- Text · Image
- License
- —
CyScenarioBench 76.3%, CVE Rediscovery 89.6%; verified defensive and research use only
G
Google DeepMind
Released
Rolled out globally Nov 12, 2026 across the Gemini API, AI Studio, the Gemini app, Search and Antigravity
Version4 Hydrogen
- Parameters
- Not disclosed
- Context window
- 1M
- Modality
- Text · Image · Video · Audio
- License
- —
- API
- Web
- Reasoning
- Functions
- Caching
AA Intelligence Index 51 (default high) / 54 (xhigh) vs Argon 50; Terminal-Bench 4.1 57% vs 18.2% for 3.8 Flash