DeepSeek ties Anthropic

Parity at 5.8

DeepSeek's share of voice among tagged developer stories on Hacker News and Lobsters was 0.3% in March. In August, through 12 days, it is 5.8%. That figure ties Anthropic (5.8%) and sits ahead of Gemini (2.6%). Claude remains dominant at 20.5%, with OpenAI at 11.2% and GPT at 8.6%.

The monthly trajectory is a steady climb, not a spike. DeepSeek's SoV moved from 0.9% in February to 0.3% in March, then 1.2% in April, 2.2% in May, 1.8% in June and 1.7% in July before reaching 5.8% in the partial August window. The jump from July to August is the largest single-month move in the series.

The crossover with Anthropic is the headline. Anthropic's August SoV of 5.8% is its lowest monthly figure since February, when it stood at 7.9%. DeepSeek has reached the same level from below.

The ramp, not the flash

The surge is built on three consecutive product launches, not a single news event. DeepSeek V4 Flash 0731 landed on July 31 with a peak HN score of 787 and 470 comments. DeepSeek V4 Pro 0813 followed on August 12, reaching a peak score of 1,810 with 1,412 comments. DeepSeek Harness, a developer preview, appeared on August 13 at a peak of 729 with 305 comments.

The weekly data shows a six-week ramp with accelerating engagement. From the week of July 6 through the week of August 10, DeepSeek's weekly story count went 10, 7, 12, 24, 26, 39. Average peak scores rose from 16 to 152. The maximum peak score in the final week reached 1,810.

This is the inverse of the Kimi K3 pattern. Kimi's weekly story count went 2, 40, 40, 54, 12, 4 across the same window. It spiked to 54 stories in the week of July 27 and collapsed to 4 by August 10. DeepSeek ended that week at 24 stories and kept climbing.

No other open-weight model exceeded 10 stories in any week of the window. Llama held at 2 to 5. Qwen ranged from 2 to 10. Mistral never exceeded 7. The open-weight conversation has narrowed to a single name.

Anthropic's descent

Anthropic's decline created the gap DeepSeek filled. Anthropic's weekly SoV peaked at 14.6% in the week of June 8, driven by its S-1 filing, Fable and Mythos model launches, the Alibaba extraction claim and the Amazon and US-officials crackdown story. By the week of August 10, Anthropic had fallen to 5.0%.

Part of the July decline came from negative narratives. The Zig creator's critique reached a peak score of 1,523. The $1.5 billion piracy settlement added pressure. These were stories about Anthropic, but they weighed on its share rather than lifting it.

DeepSeek crossed Anthropic for the first time in the week of August 10, at 6.9% vs. 5.0%. The weekly series shows the convergence: Anthropic moved from 14.6% to 5.0% over 11 weeks, while DeepSeek moved from 2.2% to 6.9%.

Zero hiring signal

The conversation surge has not produced hiring signal. DeepSeek's job_mentions_7d has been 0 in 10 of 11 weekly snapshots since June 12, reaching 1 only twice. The latest snapshot, captured August 17, shows DeepSeek at 0. GPT sits at 11, OpenAI at 10, Gemini at 9 and Claude at 8.

DeepSeek has no tracked GitHub artifact, no npm or PyPI registry presence and no Docker Hub artifacts. Entity_stats contains 12 rows for DeepSeek, all job_mentions_7d. There are no github_stars, npm_downloads, pypi_downloads, new_repos_weekly or dockerhub_downloads rows. The developer economy is talking about DeepSeek's models. It is not building on them or hiring for them, at least not in ways the tracked registries can see.

This is the divergence at the center of the story. A model can dominate developer conversation without a commercial ecosystem. Whether that situation holds is the question the next two months will answer.

What this confirms

Issue Nº 0 (June 12) asked whether Codex would consolidate the number-two position in the AI coding-tool race. It has. Codex held 30 to 52 weekly stories through July and August, vs. Cursor's 9 to 15. The crossing noted in June is now a structural gap.

Issue Nº 3 tracked MCP adoption against the bet that its monthly SoV would stay at or below its March peak of 9.0% through Q3. MCP's July SoV was 8.6%. Its new_repos_weekly rose from 1,258 (June 22) to 1,850 (August 17), a 47% increase over eight weeks. The plumbing is growing; the argument is not returning.

Issue Nº 4 will cover Kimi K3's spike-and-collapse pattern. This issue is the complementary story: where Kimi flashed and faded, DeepSeek ramped and held. The two are the archetypes of challenger-model mindshare in the summer of 2026.