Foundation-Model on Best of AI

abab6.5

Thu, 16 Jul 2026 00:00:00 +0000

abab6.5 is a large language model from the Chinese company MiniMax, launched in April 2024 as the company’s flagship offering before it shifted toward open-weight releases with the M-series models. It runs on a mixture-of-experts architecture with a total parameter count in the trillions, though MiniMax has not disclosed exact figures, and it handles a 200,000-token context window. The lineup also included a companion model, abab6.5s, tuned for lower latency and cheaper inference at similar quality.

ALBERT

Thu, 16 Jul 2026 00:00:00 +0000

ALBERT (A Lite BERT) is a language representation model that Google Research and Toyota Technological Institute published in September 2019 as a follow-up to BERT. It uses two parameter-reduction tricks, factorized embedding parameterization and cross-layer parameter sharing, to cut the parameter count sharply while keeping accuracy competitive with much larger models. The largest configuration, ALBERT-xxlarge at around 223 million parameters, has roughly 18 times fewer parameters than BERT-large yet trains faster and set new state-of-the-art scores on GLUE, SQuAD 2.0, and RACE at the time of release.

Apple Foundation Model

Thu, 16 Jul 2026 00:00:00 +0000

Apple’s Foundation Models are the pair of language models behind Apple Intelligence, introduced in June 2024 at WWDC: a roughly 3-billion-parameter model that runs directly on iPhone, iPad, and Mac hardware, and a larger model that runs on Apple’s own Private Cloud Compute servers for tasks the on-device model can’t handle alone. The on-device model relies on techniques like KV-cache sharing and 2-bit quantization-aware training to fit within a phone’s memory and power budget, while the server model uses a mixture-of-experts transformer design. The on-device model’s usable context window is limited to 4,096 tokens, while the server model handles up to 32,000 tokens.

Baichuan 3

Thu, 16 Jul 2026 00:00:00 +0000

Baichuan 3 is a closed, API-only foundation model that Baichuan Intelligent Technology launched in January 2024, marking the point where the company stopped open-sourcing its flagship models and moved to a proprietary hosted service instead. Baichuan hasn’t disclosed the parameter count, though the company has said it improved on tasks like Chinese-language reasoning, code generation, and math over Baichuan 2, and claimed wins over GPT-4 on some Chinese benchmarks such as CMMLU and GAOKAO-style evaluations. It supports a 128K token context window and is accessed only through Baichuan’s API and chat product rather than downloadable weights. It sits between the open Baichuan 2 line and the more capable Baichuan 4 that followed later in 2024, and it was aimed mainly at enterprise customers inside China rather than at competing head-to-head with Western frontier models internationally.

Baichuan 4

Thu, 16 Jul 2026 00:00:00 +0000

Baichuan 4 is Baichuan Intelligent Technology’s flagship proprietary language model, with its API opened to developers in mid-2024. Like its predecessor Baichuan 3, the company hasn’t published a parameter count, but it has positioned the model around knowledge retrieval, long-context handling, and Chinese-language generation, and reported leading results on the domestic SuperCLUE benchmark. The family includes a Turbo variant tuned for enterprise deployment and an Air variant described as a mixture-of-experts model built for lower-cost inference, with Baichuan claiming the Turbo model priced at around 80% of GPT-4o’s API cost while offering comparable or better usability on Chinese tasks. It’s available only through Baichuan’s API and chat interface, with context windows reported up to 128K tokens in most deployments. Baichuan 4 competes mainly against other Chinese closed models like those from Alibaba’s Qwen team, Zhipu, and Moonshot rather than against Western frontier labs.

Claude 1

Thu, 16 Jul 2026 00:00:00 +0000

Claude 1 is Anthropic’s first publicly released language model, launched in March 2023 alongside the smaller, faster Claude Instant. It came out a few months after Anthropic first previewed Claude to a small group of partner companies in late 2022, and it made Anthropic’s Constitutional AI training approach, using AI-generated feedback guided by a written set of principles instead of relying only on human raters, available through a commercial API for the first time. Claude 1 launched with a 9,000-token context window, small by later standards, though Anthropic expanded it to 100,000 tokens just two months after release, well ahead of most competing chat models at the time.

Claude 2

Thu, 16 Jul 2026 00:00:00 +0000

Claude 2 is the second major Claude model generation Anthropic released, launching in July 2023 as a broad upgrade over Claude 1 in reasoning, coding, and factual accuracy. It extended the context window to 100,000 tokens as standard, letting it process documents such as long contracts, codebases, or entire books in a single request, and Anthropic reported an MMLU score around 78.5%, a notable jump for the time. Claude 2 was also the first Claude model made available directly to consumers through a standalone chat interface at claude.ai in the United States and United Kingdom, rather than only through the API and partner products.

Claude 3 Haiku

Thu, 16 Jul 2026 00:00:00 +0000

Claude 3 Haiku is the smallest and fastest model in Anthropic’s original Claude 3 family, launched in March 2024 alongside Sonnet and Opus. It was built for high-volume, latency-sensitive work such as customer support chat, content moderation, and data extraction, where a slower flagship model would be overkill. At $0.25 per million input tokens and $1.25 per million output tokens, it undercut most competing models on price while still reading images alongside text and handling a 200,000 token context window. Anthropic has since released faster, cheaper models in the Haiku line, so Claude 3 Haiku is now a legacy option kept available mainly for existing integrations built on it.

Claude 3 Opus

Thu, 16 Jul 2026 00:00:00 +0000

Claude 3 Opus was Anthropic’s flagship model when the Claude 3 family launched in March 2024, sitting above Sonnet and Haiku as the largest and most capable of the three. It handled complex analysis, long documents, and multi-step tasks better than any earlier Claude model, and at launch it beat GPT-4 on several public benchmarks, which helped establish Anthropic as a serious frontier lab rather than a distant second to OpenAI. It reads both text and images, supports a 200,000 token context window, and is priced at $15 per million input tokens and $75 per million output tokens, well above the Sonnet and Haiku tiers released alongside it. Anthropic has since shipped several generations of stronger, cheaper models, so Claude 3 Opus is now considered a legacy model rather than a current recommendation.

Claude 3 Sonnet

Thu, 16 Jul 2026 00:00:00 +0000

Claude 3 Sonnet is the middle tier of Anthropic’s original Claude 3 family, released in March 2024 alongside the smaller Haiku and the larger Opus. It was pitched as the balance point of the lineup: strong enough for most business tasks like summarization, coding help, and data analysis, but priced and sized to run more cheaply than Opus. At $3 per million input tokens and $15 per million output tokens, with a 200,000 token context window and support for image inputs, it became a common default for teams building on the Claude API in 2024. Anthropic replaced it a few months later with Claude 3.5 Sonnet, which outperformed it on most benchmarks at the same price, so Claude 3 Sonnet is now a legacy model.

Claude 3.5 Haiku

Thu, 16 Jul 2026 00:00:00 +0000

Claude 3.5 Haiku is Anthropic’s fast, low-cost model in the Claude 3.5 family, released in October 2024 alongside an upgraded Claude 3.5 Sonnet. Anthropic built it to replace Claude 3 Haiku as the cheapest, quickest option in the lineup while closing much of the capability gap with larger models: on coding evaluations such as SWE-bench Verified, Claude 3.5 Haiku scored competitively with the original Claude 3.5 Sonnet and outperformed several larger models available at launch, and Anthropic reported an 88.1% score on HumanEval for code generation.

Claude 3.5 Sonnet

Thu, 16 Jul 2026 00:00:00 +0000

Claude 3.5 Sonnet is a text-and-image language model from Anthropic, first released in June 2024 as a mid-tier upgrade to the Claude 3 lineup that matched or beat the larger Claude 3 Opus on most benchmarks while running faster and cheaper. Anthropic updated it again in October 2024 with a version that added computer use, a feature letting the model view a screenshot and control a mouse and keyboard to operate software the way a person would. That update made Claude 3.5 Sonnet one of the first widely available models built for agentic tasks like filling out web forms, navigating operating systems, and running multi-step workflows without a human clicking each button. It kept the 200,000 token context window of its predecessors and priced at $3 per million input tokens and $15 per million output tokens, positioning it as Anthropic’s workhorse model through much of 2024 and early 2025 before the Claude 3.7 and Claude 4 lines replaced it.

Claude Haiku 4.5

Thu, 16 Jul 2026 00:00:00 +0000

Claude Haiku 4.5 is Anthropic’s fastest and cheapest current-generation model, released on October 15, 2025. It is the first Haiku model to include extended thinking, computer use, and context awareness, features that had previously only shipped in Anthropic’s larger Sonnet and Opus models. On SWE-bench Verified it scores about 73.3 percent, which puts it within a few points of Claude Sonnet 4.5’s 77.2 percent despite costing a third as much: $1 per million input tokens and $5 per million output tokens versus Sonnet’s $3/$15. That makes it a practical choice for running many agents in parallel or handling high-volume tasks where Sonnet or Opus would be needlessly expensive. It reads both text and images and shares the 200,000 token context window used across the rest of the current Claude lineup.

Claude Sonnet 4

Thu, 16 Jul 2026 00:00:00 +0000

Claude Sonnet 4 launched on May 22, 2025 alongside the larger Claude Opus 4, taking the mid-tier spot in Anthropic’s Claude 4 generation. It is a hybrid reasoning model, meaning it can answer directly or work through a problem step by step in an extended thinking mode depending on the task, and Anthropic reported it hitting roughly 72.7 percent on SWE-bench Verified at launch, edging out even Opus 4 on that particular benchmark while costing a fraction of the price. It ships with a 200,000 token context window as standard, with a 1 million token context window available in beta on the API and Amazon Bedrock, and it reads images alongside text. At $3 per million input tokens and $15 per million output tokens, it became Anthropic’s default recommendation for coding and everyday agentic work where Opus-level cost isn’t justified.

Claude Sonnet 5

Thu, 16 Jul 2026 00:00:00 +0000

Claude Sonnet 5 launched on June 30, 2026 as Anthropic’s most agentic Sonnet model yet, built to plan multi-step tasks and use tools like browsers and terminals with less hand-holding than earlier Sonnet versions. It performs close to Claude Opus 4.8 on most benchmarks, including roughly 85.2 percent on SWE-bench Verified and 63.2 percent on SWE-bench Pro, while costing noticeably less, which is why Anthropic and reviewers have framed it as a cheaper way to run coding agents at scale. It carries a 1 million token context window with context compaction to manage long agent sessions, and reads both text and images. Anthropic priced it at an introductory $2 per million input tokens and $10 per million output tokens through August 31, 2026, after which it moves to standard pricing of $3/$15, and it is now the default model for free and Pro Claude.ai plans.

CLIP

Thu, 16 Jul 2026 00:00:00 +0000

CLIP (Contrastive Language-Image Pre-training) is a model OpenAI released in January 2021 that learns to match images with text descriptions. It was trained on 400 million image-text pairs scraped from the internet, using a contrastive objective that pulls matching image and text embeddings together while pushing mismatched pairs apart. The largest released variant, built on a ViT-L/14 vision transformer paired with a 63M-parameter text transformer, totals roughly 428 million parameters. Instead of being trained for one narrow task, CLIP learns a general-purpose joint embedding space for images and text, which lets it perform zero-shot image classification by comparing an image against a set of text labels with no task-specific fine-tuning.

Codex

Thu, 16 Jul 2026 00:00:00 +0000

Codex is OpenAI’s first dedicated code-generation model, announced in August 2021 as a GPT-3 descendant fine-tuned on billions of lines of public source code from GitHub. The largest published version had 12 billion parameters and a 4,096-token context window, and it solved 28.8% of problems on OpenAI’s HumanEval benchmark on the first try, rising to 77.5% when allowed 100 sampled attempts per problem. Codex is what powered the original GitHub Copilot when Copilot launched in technical preview in 2021, marking one of the first large-scale deployments of an LLM as a coding assistant rather than a chat or completion demo. OpenAI made it available through a private beta API for a few years before retiring the standalone Codex API in March 2023, folding its capabilities into successor GPT-3.5 and GPT-4 family models. The name lives on today as the brand for OpenAI’s newer agentic coding products, but the original Codex model itself is no longer accessible through the API.

ELECTRA

Thu, 16 Jul 2026 00:00:00 +0000

ELMo

Thu, 16 Jul 2026 00:00:00 +0000

ERNIE 4.0

Thu, 16 Jul 2026 00:00:00 +0000

Gemini 3 Pro

Thu, 16 Jul 2026 00:00:00 +0000

GLM-4-Plus

Thu, 16 Jul 2026 00:00:00 +0000

GPT-1

Thu, 16 Jul 2026 00:00:00 +0000

GPT-3

Thu, 16 Jul 2026 00:00:00 +0000

GPT-3 Davinci

Thu, 16 Jul 2026 00:00:00 +0000

GPT-3.5 Turbo

Thu, 16 Jul 2026 00:00:00 +0000

GPT-4

Thu, 16 Jul 2026 00:00:00 +0000

GPT-4 Turbo

Thu, 16 Jul 2026 00:00:00 +0000

GPT-4.5

Thu, 16 Jul 2026 00:00:00 +0000

GPT-4o

Thu, 16 Jul 2026 00:00:00 +0000

GPT-5

Thu, 16 Jul 2026 00:00:00 +0000

GPT-5.6 Terra

Thu, 16 Jul 2026 00:00:00 +0000

Grok 3

Thu, 16 Jul 2026 00:00:00 +0000

Grok 4

Thu, 16 Jul 2026 00:00:00 +0000

Grok-2

Thu, 16 Jul 2026 00:00:00 +0000

HyperCLOVA X

Thu, 16 Jul 2026 00:00:00 +0000

Inflection-2.5

Thu, 16 Jul 2026 00:00:00 +0000

InstructGPT

Thu, 16 Jul 2026 00:00:00 +0000

Jurassic-1

Thu, 16 Jul 2026 00:00:00 +0000

Jurassic-2

Thu, 16 Jul 2026 00:00:00 +0000

Jurassic-2 Mid

Thu, 16 Jul 2026 00:00:00 +0000

Llama 3.1 405B

Thu, 16 Jul 2026 00:00:00 +0000

Llama 3.1 70B

Thu, 16 Jul 2026 00:00:00 +0000

Llama 3.3

Thu, 16 Jul 2026 00:00:00 +0000

Mistral Large 2

Thu, 16 Jul 2026 00:00:00 +0000

Mistral Large 3

Thu, 16 Jul 2026 00:00:00 +0000

Mistral Medium 3

Thu, 16 Jul 2026 00:00:00 +0000

PaLM

Thu, 16 Jul 2026 00:00:00 +0000

PaLM 2

Thu, 16 Jul 2026 00:00:00 +0000

Qwen-7B

Thu, 16 Jul 2026 00:00:00 +0000

Qwen-Max

Thu, 16 Jul 2026 00:00:00 +0000

Qwen2.5-72B

Thu, 16 Jul 2026 00:00:00 +0000

Qwen2.5-Max

Thu, 16 Jul 2026 00:00:00 +0000

RoBERTa

Thu, 16 Jul 2026 00:00:00 +0000

Samsung Gauss 2

Thu, 16 Jul 2026 00:00:00 +0000

SenseChat 5

Thu, 16 Jul 2026 00:00:00 +0000

Spark 4.0

Thu, 16 Jul 2026 00:00:00 +0000

Step-1

Thu, 16 Jul 2026 00:00:00 +0000

text-davinci-003

Thu, 16 Jul 2026 00:00:00 +0000

XLNet

Thu, 16 Jul 2026 00:00:00 +0000

Yi-Large

Thu, 16 Jul 2026 00:00:00 +0000