Conversation vs. Adoption · Issue Nº 7
Chinese models took 12% of AI-model conversation and zero adoption
DeepSeek and Qwen grew 5.7× in developer mindshare from February to August while leaving no trace in package registries or job postings.
February 1 – August 26, 2026 · Published 2026-08-26
The tie that wasn't
In Issue Nº 5 (Aug 22), DeepSeek appeared to tie Anthropic at 5.8% each on 12 days of August data. With 26 days counted, the tie is over. DeepSeek reached 8.9% (110 stories) against Anthropic's 5.2% (138 stories), a 70% lead. DeepSeek's monthly share among ai_model entities grew 5.2× from February's 1.7%.
The open-weight conversation has also stopped being a single name. Issue Nº 5 noted that "no other open-weight model exceeded 10 stories in any week" and that the conversation had "narrowed to a single name." Qwen's 37 August stories, 4 of which reached the front page, invalidate that claim. The Chinese open-weight conversation is now a duo.
The monthly series for combined Chinese share reads: 2.1% (Feb), 1.8% (Mar), 2.9% (Apr), 5.0% (May), 4.4% (Jun), 4.4% (Jul), 11.9% (Aug, partial). The August figure is a floor, not a ceiling, since it covers 26 of 31 days.
| February | 1.7 | 0.4 | 2.1 |
|---|---|---|---|
| March | 1.5 | 0.3 | 1.8 |
| April | 2.5 | 0.4 | 2.9 |
| May | 4.4 | 0.6 | 5.0 |
| June | 3.9 | 0.5 | 4.4 |
| July | 3.7 | 1.5 | 4.4 |
| August (partial) | 8.9 | 3.0 | 11.9 |
The cadence
The surge was not a single product launch. In a three-week window, DeepSeek shipped V4 Flash 0731 (Aug 7, peak 184), V4 Pro 0813 (Aug 12, peak 1,031, 446 comments), Harness developer preview (Aug 13, peak 729, 305 comments), and V4 flash-vision-exp (Aug 21, peak 496). Qwen shipped 3.8 27B on Aug 14 (peak 1,425), out-scoring every DeepSeek story except V4 Pro.
The weekly series confirms a sustained ramp, not a one-week spike. Combined DeepSeek and Qwen stories went 12, 14, 20, 31, 35, 57, 31 (partial) across weeks of Jul 6 through Aug 24. The acceleration began the week of July 27 with the V4 Flash 0731 launch and intensified through August. In the peak week (Aug 10), 57 Chinese-model stories appeared against Anthropic's 186, a ratio of 1:3.3. In early June that ratio was 1:15.
The conversation is broad-based. DeepSeek's 110 August stories came primarily from hn_new (99), but 11 reached hn_top. Qwen's 37 stories came from hn_new (31) and hn_top (4). One story each reached hn_show and lobsters for Qwen; one reached hn_show for DeepSeek. The 15 front-page placements confirm these are not just noise from new submissions.
| Week of Jul 6 | 12 | — | — |
|---|---|---|---|
| Week of Jul 13 | 14 | — | — |
| Week of Jul 20 | 20 | — | — |
| Week of Jul 27 | 31 | — | — |
| Week of Aug 3 | 35 | — | — |
| Week of Aug 10 | 57 | 186 | 1:3.3 |
| Week of Aug 17 | 31 | — | — |
| Week of Aug 24 (partial) | 15 | 62 | — |
The adoption void
The entity_stats table for DeepSeek and Qwen contains only one metric: job_mentions_7d. No github_stars, npm_downloads, pypi_downloads, dockerhub_pulls, or new_repos_weekly rows exist for either entity. DeepSeek's job_mentions_7d was 0 in 10 of 11 weekly snapshots since June 12, reaching 1 only twice (Aug 3, Aug 10). Qwen's was 0 in 8 of 11 snapshots, reaching 1 or 2 only three times.
The Western models developers are actually installing grew. Anthropic npm downloads rose 47.5% to 34.8M/week. OpenAI npm rose 38.9% to 35.5M/week. Claude Code npm rose 106.5% to 22.8M/week. Anthropic PyPI rose 50.9% to 47.3M. OpenAI PyPI rose 37.7% to 97.4M.
The divergence is total. Conversation share rose 5.7× while adoption is literally zero, not just slow. Developers are downloading from Anthropic and OpenAI while talking about DeepSeek and Qwen. This is the extreme case of the pattern Issue Nº 6 documented for Anthropic, where conversation share fell while downloads rose. The Chinese models invert it: conversation rising while nothing gets installed.
The adoption gap is structural, not a timing lag. These models are consumed via API calls and self-hosted weights, not package registries. Eleven weeks of tracking have produced no registry artifacts for either entity, and there is no mechanism by which they would. The conversation will continue to grow; the download counters will not move.
| Anthropic npm | 23.6M | 34.8M | +47.5 |
|---|---|---|---|
| OpenAI npm | 25.5M | 35.5M | +38.9 |
| Claude Code npm | 11.1M | 22.8M | +106.5 |
| Anthropic PyPI | 31.4M | 47.3M | +50.9 |
| OpenAI PyPI | 70.7M | 97.4M | +37.7 |
| DeepSeek (all registries) | 0 | 0 | — |
| Qwen (all registries) | 0 | 0 | — |
Candidate bets
DeepSeek + Qwen combined will exceed 15% of AI-model conversation share in September 2026. August hit 11.9% on a partial month with a three-week launch cadence. Qwen's 10.8% front-page rate suggests its stories have outsized staying power. If DeepSeek ships one more V4 variant and Qwen 3.8 sees follow-on coverage, 15% is reachable. Falsifiable: combined share below 15% for September.
Qwen's monthly AI-model conversation share will exceed DeepSeek's by December 2026. Qwen's 10.8% front-page rate is 2.4× DeepSeek's, meaning its stories resonate more per mention. Qwen also has a commercial ecosystem (Alibaba) that could eventually produce tracked artifacts. DeepSeek has no tracked artifacts and no commercial ecosystem. The gap (8.9% vs 3.0%) is large but the trajectory is closing. Falsifiable: Qwen share below DeepSeek share in any month through December.
DeepSeek and Qwen will have zero tracked GitHub, npm, or PyPI artifacts through October 31, 2026. Neither entity has any registry presence after 11 weeks of tracking. Their models are consumed via API or self-hosted weights, not package registries. The adoption gap is structural, not a timing lag. Falsifiable: any tracked artifact appears for either entity.