Skip to content
AImpact
IT EN
← Reading paths

Reading path

Open-source developer in the model era

Llama, Mistral, Gemma, DeepSeek: the history of open weights that matter.

You are a developer who does not want to depend on closed APIs and believes in the value of open-weight models: you can inspect the weights, fine-tune, and run everything locally. This path follows the open community's history — from EleutherAI and GPT-Neo through Llama 4 and DeepSeek — covering the milestones that redefined what you can do without paying tokens to anyone.

  1. 01

    Why it matters to you

    EleutherAI's first large open-weight model: proof that open research could challenge OpenAI at scale, before Hugging Face became the center of the world.

    High Open Source Models

    GPT-Neo: the first open source clone of GPT-3

    EleutherAI releases GPT-Neo 1.3B and 2.7B, open source language models trained on The Pile — the first serious attempt to replicate the GPT-3 architecture with public weights.

  2. 02

    Why it matters to you

    176 billion parameters, trained collaboratively by hundreds of researchers: the model that showed a distributed community can compete with private labs.

    High Open Source Models

    BLOOM 176B: the first truly open large multilingual LLM

    The BigScience collective releases BLOOM, a 176-billion-parameter model trained on 46 human languages and 13 programming languages, under an open RAIL license.

  3. 03

    Why it matters to you

    The leak that changed everything: LLaMA weights circulate freely and within weeks anyone can fine-tune on a laptop. The modern open ecosystem is born.

    High Open Source Models

    LLaMA: Meta opens foundation models to research

    Meta releases LLaMA in four sizes (7B, 13B, 33B, 65B), available to researchers on request. One week later, the weights leak publicly.

  4. 04

    Why it matters to you

    7 billion parameters that beat Llama 2 13B on almost every benchmark: Mistral proves that efficiency and architecture matter more than raw scale.

    High Open Source Models

    Mistral 7B: Europe joins the open-source race

    Mistral AI (Paris), a three-month-old startup founded by ex-Meta/DeepMind researchers, releases Mistral 7B under Apache 2.0. Beats Llama 2 13B on most benchmarks with half the parameters.

  5. 05

    Why it matters to you

    Mixture-of-Experts open to the community: the technique that delivers 70B-quality at 13B inference cost becomes accessible to anyone with a serious GPU.

    Landmark Open Source Models

    Mixtral 8x7B: open-source Mixture of Experts that beats GPT-3.5

    Mistral drops Mixtral 8x7B via magnet link with no warning: SMoE with 8 experts of 7B, 13B active params out of 47B total. Performance matches/exceeds GPT-3.5. Apache 2.0.

  6. 06

    Why it matters to you

    Google releases open weights optimized for single-GPU deployment: the signal that even big labs must reckon with the open-weight ecosystem.

    High Open Source Models

    Gemma: Google enters the open-weights game

    Google releases Gemma 2B and 7B, open-weight models derived from Gemini research. For the first time Google competes directly with Llama and Mistral on open ground.

  7. 07

    Why it matters to you

    An open-weight model with reasoning competitive with OpenAI o1, trained at a fraction of the cost: the strongest proof yet that open source has reached the frontier level.

    Landmark Open Source Models

    DeepSeek-R1: open reasoning matches o1 at 1/30 the cost

    Chinese startup DeepSeek releases R1, a reasoning model with MIT-licensed open weights. Performance on par with OpenAI o1, API pricing $0.55/$2.19 per 1M tokens (vs o1 $15/$60). Nasdaq AI loses $1T in two days.