Pins matching tag:"Frontier Model"

OpenAI unveils GPT-4 Turbo at its first DevDay: a preview of the next GPT-4, gpt-4-1106-preview, with a 128K context window, knowledge to April 2023 and prices of $10 per 1 million input tokens and $30 per 1 million output tokens.

OpenAI releases gpt-3.5-turbo-0125, a GPT-3.5 Turbo update that cuts input prices 50% to $0.50 per 1 million tokens and output prices 25% to $1.50 per 1 million, with better format-following and a fix for non-English function calls.

OpenAI releases GPT-4o, its flagship model that reasons across audio, vision and text in real time, and makes it the new default in ChatGPT.

OpenAI introduces GPT-4o mini, its most capable and cost-efficient small model, surpassing GPT-3.5 Turbo on academic benchmarks.

Meta releases Llama 3.1 in 8B, 70B and 405B sizes with a 128K context window; the 405B, trained on over 15 trillion tokens, is billed as the first frontier-level open model and is free to download.

OpenAI releases o1-preview and o1-mini, the first of a new series of models that spend more time thinking before they respond.

Google releases Gemma 3, open models in 1B, 4B, 12B and 27B sizes with vision, a 128K context window and 140-language support; the 27B scores 1338 on LMArena and the 4B is pitched against Gemma 2 27B.

Alibaba's Qwen team open-sources the Qwen3 family under Apache 2.0, six dense models from 0.6B to 32B and two mixture-of-experts models (30B-A3B and 235B-A22B), with switchable thinking and non-thinking modes and 119 languages; the dense flagship Qwen3-32B scores 93.8 on Arena-Hard in thinking mode.

Alibaba's Qwen team quietly releases Qwen3-235B-A22B-Instruct-2507, a non-thinking 235B mixture-of-experts model with a 256K context under Apache 2.0 that Qwen's own tables put ahead of Claude Opus 4 non-thinking and Kimi K2 on several tests.

Alibaba's Qwen team releases Qwen3-30B-A3B-Instruct-2507, a non-thinking 30.5B mixture-of-experts model with only 3.3B active parameters and a 262K context that Qwen says approaches GPT-4o.

OpenAI releases two open-weight reasoning models, gpt-oss-120b and gpt-oss-20b, for teams that want to run and customize models themselves.

Alibaba releases Qwen3-Next-80B-A3B-Instruct and Thinking, ultra-sparse 80B models that activate only 3B parameters per token, run a hybrid Gated DeltaNet attention design and carry an Apache 2.0 licence.

DeepSeek releases the experimental DeepSeek-V3.2-Exp, the first model with DeepSeek Sparse Attention, and cuts its API prices by more than half to $0.28 per million input tokens and $0.42 per million output tokens.

Z.ai (Zhipu AI) releases GLM-4.6, its flagship coding model with a 200K-token context window, which wins 48.6% of head-to-head real-world coding tasks against Claude Sonnet 4 and uses over 30% fewer tokens than GLM-4.5.

Google releases Gemini 3 Flash, frontier intelligence built for speed that outscores 2.5 Pro at three times the speed, priced at $0.50 per million input tokens and $3 per million output tokens.

Z.ai releases GLM-4.7-Flash, a 30B-A3B mixture-of-experts open-weight model that beats gpt-oss-20b and Qwen3-30B-A3B-Thinking on most coding and agent benchmarks, with a free API tier.

DeepSeek releases the DeepSeek-V4 preview as open weights: V4-Pro with 1.6 trillion parameters (49 billion active) and V4-Flash with 284 billion (13 billion active), both with a 1 million token context window and launch API prices from $0.14 per million input tokens.

Tencent officially releases Hy3, the production version of its open-weight Hunyuan model: a 295B-parameter mixture-of-experts with 21B active parameters and a 256K-token context, priced at 1 yuan per 1 million input tokens.

IBM releases Granite 4.2, three open Apache 2.0 reasoning models (3B, 8B and 30B) with a switchable thinking mode; the 8B scores 47.67 on SWE-bench Verified and 86.67 on AIME25 in IBM's own runs.

Z.ai releases GLM-5.3-Flash, a 320B-parameter (18B active) natively multimodal open-weight model on a newly trained base that it says beats GLM-5.2 at one-tenth of the price, listed at $0.15 per million input tokens.

Google launches Gemini 3.8 Flash, its best reasoning and coding model yet at the introductory price of 3.7 Flash, alongside Gemini 3.8 Flash Cyber, a vulnerability-finding and patching model offered only to trusted government and enterprise defenders through its new Fairwind Program.

Anthropic releases Claude Opus 5.5, the first model in a new Claude 5.5 family, which it says matches Claude Fable 5.1 on most work while costing 40% less to run than Opus 5.

Xiaomi releases and open-sources MiMo-V2.6-Pro and MiMo-V2.6-Flash, natively omnimodal models with 1M-token context; Pro scores 46.32 on the Artificial Analysis Intelligence Index, the highest of any open-source model.