Chatter
Story archive
MISTRAL’S ‘LE CHONK’ GOES LARGE, POST SAYS
SunnyAIDock reports the AI firm plans to release the model’s weights by the end of October, allowing private deployment.
BEAR WITH US: Qwen’s “big or small?” teaser fuels model rumors
A poster floats Qwen4-27B and Qwen4-Max as possible open-weight releases—but says there’s no date or benchmark yet.
Claude Rumor Mill Gives GLM a Cameo
A post casts Zhipu’s open-weight GLM-5.2 as a rival amid bets on a future Claude release — not as the subject of a new launch report.
A DOUBLE WEIGHTING GAME
A post says Qwen’s “let’s do both” reply points to a simultaneous open-weight drop for Qwen4-27B and Qwen4-Max — but gives no release date.
MINIMAX’S PHYSICS BOAST MEETS THE REAL WORLD
The AI developer calls H3 the best open-source video world model for physics; a post reporting exam results puts it behind Seedance 2.5 in a 40-task test.
2026-10-18
DEEPSEEK: PENNY WISE, HOUR FOOLISH
A post reports a DeepSeek AI model cost about 23 cents on one test versus $6 for Opus — but took around an hour to Opus’ 20 minutes.
2026-10-13
MISTRAL’S ‘LE CHONK’ GOES LARGE, POST SAYS
SunnyAIDock reports the AI firm plans to release the model’s weights by the end of October, allowing private deployment.
2026-10-12
BEAR WITH US: Qwen’s “big or small?” teaser fuels model rumors
A poster floats Qwen4-27B and Qwen4-Max as possible open-weight releases—but says there’s no date or benchmark yet.
2026-10-12
Claude Rumor Mill Gives GLM a Cameo
A post casts Zhipu’s open-weight GLM-5.2 as a rival amid bets on a future Claude release — not as the subject of a new launch report.
2026-10-11
A DOUBLE WEIGHTING GAME
A post says Qwen’s “let’s do both” reply points to a simultaneous open-weight drop for Qwen4-27B and Qwen4-Max — but gives no release date.
2026-10-09
MINIMAX’S PHYSICS BOAST MEETS THE REAL WORLD
The AI developer calls H3 the best open-source video world model for physics; a post reporting exam results puts it behind Seedance 2.5 in a 40-task test.
2026-10-09
TOKEN IT SLOW
STACCoverflow says its Qwen3-8B chat runs on a single-validator Solana fork — and a short exchange takes about 20–30 minutes.
2026-10-08
Deep Pockets for DeepSeek? Post Reports $12B-Plus Raise
CATL and Tencent are named among the biggest committers, with a close expected in October, according to the post.
2026-10-07
Chip off the old bot? Post says agent-led accelerator runs Qwen
A post relays the openTPU developer’s reported FPGA test of Qwen3-0.6B at about 21.6 tokens a second using int8 weights.