←Search results

Search “Reasoning”

By relevance · Total 199 results Clear search

Sunday

  1. Story1 reports · 1 facts

    Fireworks evaluates DeepSeek's new model

    On September 14, 2026, Fireworks AI released and evaluated DeepSeek-V4.1-Flash. According to its published results, the model achieves 74.34% pass@1 on DeepSWE, on par with GPT-6-Astra's 74.12%; its cost per task is $0.43, about 1/15 that of Astra. The model is a 552B-parameter MoE, using separate activation budgets of 8B for input and 16B for output, and its KV Cache HBM footprint is reduced to 1/4 that of the previous-generation V4. On HLE it scores 34.52%, below Astra's 50.40%, and Fireworks considers it better suited to agentic coding than to highly difficult reasoning. On September 28, Fireworks AI released its first cache-aware router, FireRouter with Opus, which is now live in the CLI and offered for the first time as a standalone routing model; it can select among Claude Opus 5.5, GLM 5.3 and GLM 5.3 Flash based on task fit, prompt cache cost and switching cost; its internal A/B testing over more than a month showed coding task cost per session dropping from $15.36 to $6.63, a 57% reduction (±19 percentage points), with accuracy at pure O

    On September 14, 2026, Fireworks AI released and evaluated DeepSeek-V4.1-Flash. According to its published results, the model achieves 74.34% pass@1 on DeepSWE, on par with GPT-6-Astra's 74.12%; its cost per task is $0.43, about 1/15 that of Astra. The model is a 552B-parameter MoE, using separate activation budgets of 8B for input and 16B for output, and its KV Cache HBM footprint is reduced to 1/4 that of the previous-generation V4. On HLE it scores 34.52%, below Astra's 50.40%, and Fireworks

    Matched fieldsSummary

Wednesday

Tuesday

Sunday

  1. Story1 reports · 1 facts

    Ant's ASystem team launches AReno local post-training toolkit

    On August 14, 2026, Ant's ASystem team launched the AReno local large-model post-training toolkit. On the same day, Ant Ling released a practice article describing how AReno was used on DGX Spark to perform Agentic RL on Ling-3.0-tiny, using tic-tac-toe as the minimal validation task and completing a four-step closed loop of baseline evaluation, reward definition, post-training, and re-testing in the same environment. The experiment used GSPO to train for 400 steps; rollout/rewards_mean rose from about -0.5 to around 0.4, response_len dropped to about 850 tokens, and in the re-test, tool calls and move selection were more stable. Ling-3.0-tiny is a hybrid reasoning model with 7.9B total parameters and 1.3B activated parameters, and BF16, FP8, and INT4 versions have been open-sourced; AReno provides a Docker reproduction tutorial. The toolkit has now been made public together with the above practice.

    On August 14, 2026, Ant's ASystem team launched the AReno local large-model post-training toolkit. On the same day, Ant Ling released a practice article describing how AReno was used on DGX Spark to perform Agentic RL on Ling-3.0-tiny, using tic-tac-toe as the minimal validation task and completing a four-step closed loop of baseline evaluation, reward definition, post-training, and re-testing in the same environment. The experiment used GSPO to train for 400 steps; rollout/rewards mean rose fro

    Matched fieldsSummary

Friday

Tuesday

Wednesday

Wednesday

  1. Ma Dongxi NLP

    Reinforcement Learning Terminology for LLM Readers: How a Single Score Is Distributed to Each Token

    The article uses DeepSeek-R1-Zero's pass@1 on AIME 2024 rising from 15.6% to 71.0% as an example to show that a rule-based judge gives only a single score for the entire response, whereas training needs to adjust the probability of each token. The author uses the sampling example "Is OpenAI open?" to explain in turn policy, state, action, reward, return, value, Q, advantage, baseline, TD error, GAE, clipping, and KL, and points out that PPO uses a critic to learn V, while GRPO uses the average score within a group to replace V.

    The article uses DeepSeek-R1-Zero's pass@1 on AIME 2024 rising from 15.6% to 71.0% as an example to show that a rule-based judge gives only a single score for the entire response, whereas training needs to adjust the probability of each token. The author uses the sampling example "Is OpenAI open?" to explain in turn policy, state, action, reward, return, value, Q, advantage, baseline, TD error, GAE, clipping, and KL, and points out that PPO uses a critic to learn V, while GRPO uses the average s

    Matched fieldsBody

    #hot-signal#Tutorials / practice#Reasoning#English text is being prepared.

Thursday

  1. Artificial AnalysisSelected

    Gemini 4 Carbon Tops the Artificial Analysis Intelligence Index…

    Artificial Analysis evaluations show that Google's Gemini 4 Carbon tops Intelligence Index v5 with 59 points, ranking first in four of the 9 evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, and AutomationBench-AA. On the Coding Agent Index, Gemini 4 Carbon (high) paired with Antigravity CLI scores 59 points, 3 points behind Claude Code paired with Claude Opus 5.5 (max, 62 points). The model is currently available only to some users and has not yet been publicly released; it is priced at $4/$20 per 1M input/output tokens with a 50% discount, has a 1M-token context window, and supports text, image, video, and voice input.

    Artificial Analysis evaluations show that Google's Gemini 4 Carbon tops Intelligence Index v5 with 59 points, ranking first in four of the 9 evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, and AutomationBench-AA. On the Coding Agent Index, Gemini 4 Carbon (high) paired with Antigravity CLI scores 59 points, 3 points behind Claude Code paired with Claude Opus 5.5 (max, 62 points). The model is currently available only to some users and has not yet been publicly released; it is p

    Matched fieldsBody

    #editorial#Evaluations / benchmarks#Reasoning#English text is being prepared.

Friday

  1. Artificial Analysis

    Phonic 1 Tops the Artificial Analysis Speech-to-Speech Index, Leading Grok with 87.1 Points…

    Artificial Analysis evaluations show that Phonic's Phonic 1 (High) ranks first among 22 models on the Speech to Speech Index with 87.1 points, leading Grok Voice 3.0's 86.9 points by a narrow margin. The model ranks first on the Tau Voice agent task with 79.9%, leading the second place by 5.4 percentage points; it ranks third on Big Bench Audio with 98.9%, and its Time to First Audio is 0.61 seconds. Phonic 1 is available through the Phonic API, with input audio costing $0.07 per minute and $4.20 per hour.

    Artificial Analysis evaluations show that Phonic's Phonic 1 (High) ranks first among 22 models on the Speech to Speech Index with 87.1 points, leading Grok Voice 3.0's 86.9 points by a narrow margin. The model ranks first on the Tau Voice agent task with 79.9%, leading the second place by 5.4 percentage points; it ranks third on Big Bench Audio with 98.9%, and its Time to First Audio is 0.61 seconds. Phonic 1 is available through the Phonic API, with input audio costing $0.07 per minute and $4.2

    Matched fieldsBody

    #editorial#Evaluations / benchmarks#Speech#Agents